Station-city complex operation and maintenance business intelligent retrieval method and system based on large model

Abstract

The present disclosure relates to a large model-based station-city complex operation and maintenance business intelligent retrieval method and system, wherein the method comprises: obtaining user input operation and maintenance query information related to station-city complex operation and maintenance; inputting the operation and maintenance query information into the operation and maintenance large model to query and obtain corresponding operation and maintenance response information; the operation and maintenance large model is obtained by training an original large model based on sample operation and maintenance data of the station-city complex; the large model comprises a first model and a plurality of second models, and the training process comprises: iteratively training each second model independently according to the text data and the labeled data of each business type until the performance parameters of each second model meet the preset conditions to stop training; inputting the text data of each business type in the sample operation and maintenance data into each trained second model, constructing a feature set according to the output of each second model, and executing integrated training of the first model based on the feature set, and deploying the first model after training as the operation and maintenance large model.

Description

TECHNICAL FIELD

[0001] Embodiments of the present disclosure relate to the technical field of intelligent traffic management, in particular to a station-city complex operation and maintenance business intelligent retrieval method and system based on a large model. BACKGROUND

[0002] Currently, in a large traffic system, there are various types of equipment in a station-city complex, and the running state changes are complex. Conventional manual management methods and tools cannot efficiently process massive information. In the current operation and maintenance decision-making process, operation and maintenance personnel need to spend a lot of time obtaining and sorting data when working, and lack intelligent support. With the increasing complexity of equipment and systems, traditional operation and maintenance management methods have been difficult to cope with the increasingly complex operation and maintenance needs, especially as the operation and maintenance tasks continue to increase, the requirements for response speed and decision-making accuracy are also increasing. Traditional operation and maintenance methods usually rely on manual operation and experience judgment, resulting in long response time and errors in operation and maintenance management results. SUMMARY

[0003] In order to solve the above technical problems or at least partially solve the above technical problems, embodiments of the present disclosure provide a station-city complex operation and maintenance business intelligent retrieval method and system based on a large model.

[0004] In a first aspect, embodiments of the present disclosure provide a station-city complex operation and maintenance business intelligent retrieval method based on a large model, which comprises:

[0005] obtaining operation and maintenance query information related to station-city complex operation and maintenance input by a user;

[0006] inputting the operation and maintenance query information into an operation and maintenance large model to query and obtain corresponding operation and maintenance response information; wherein the operation and maintenance large model is obtained by training an original large model based on sample operation and maintenance data of a station-city complex; the original large model comprises a first model and a plurality of second models, the sample operation and maintenance data comprises a plurality of different business type text data and annotation data related to operation and maintenance; the total number of the plurality of second models is the same as the total number of the plurality of different business types;

[0007] and wherein the training process of the original large model comprises: iteratively training each second model independently according to the text data and annotation data of each business type until the performance parameters of each second model meet the preset conditions to stop training, wherein the text data of each business type corresponds to a second model; inputting the text data of each business type in the sample operation and maintenance data into each trained second model, constructing a feature set according to the output of each second model, and executing integrated training of the first model based on the feature set to deploy the first model after training as the operation and maintenance large model.

[0008] In an embodiment, the plurality of second models are pre-trained large models, and the training process of the original large model specifically includes:

[0009] Each of the pre-trained large models is independently fine-tuned according to at least part or all of the text data and the annotation data of each business type until the performance parameters of each of the pre-trained large models meet the preset condition.

[0010] Each of the pre-trained large models is input with the text data of each business type, and a feature set is constructed according to the output of each of the pre-trained large models. The first model is trained based on the feature set, and the training is ended when the performance parameters of the first model meet the preset condition. The first model after the training is deployed as the operation and maintenance large model.

[0011] In an embodiment, the training process of each second model includes:

[0012] A plurality of entity keywords and relationships between the plurality of entity keywords are identified from the text data of the sample operation and maintenance data, and a knowledge graph of a graph-based text index is constructed based on the plurality of entity keywords and the relationships between the plurality of entity keywords.

[0013] The retrieval enhancement generation model adopts a double-layer retrieval mechanism to retrieve and extract query keywords in the text data of each business type, queries target keywords similar to the query keywords from the knowledge graph, inputs the query keywords and the target keywords into the second model to obtain generated content, and updates the second model based on the difference between the generated content and the annotation content.

[0014] In an embodiment, the method further includes:

[0015] Displaying a query user interface, the query user interface including a search box for receiving user input of the operation and maintenance query information, the operation and maintenance query information including at least operation and maintenance related keywords, categories or dates.

[0016] In an embodiment, the query user interface includes a first control, and the method further includes:

[0017] In response to a triggering operation on the first control, displaying common questions and answers about station city complex operation and maintenance, or obtaining historical query records of the user and displaying questions and answers about station city complex operation and maintenance related to the historical query records.

[0018] In an embodiment, the query user interface includes a second control, and the method further includes:

[0019] In response to a triggering operation on the second control, display historical operation and maintenance query information of the user and historical operation and maintenance response information given by the operation and maintenance large model;

[0020] Obtain filtering information input by the user, and filter target operation and maintenance query information and / or target operation and maintenance response information from the historical operation and maintenance query information and the historical operation and maintenance response information based on the filtering information.

[0021] In one embodiment, the method further comprises:

[0022] Collect historical operation and maintenance query information input by the user and historical operation and maintenance response information given by the operation and maintenance large model, and obtain updated text data in the plurality of different business type text data;

[0023] Extract key feature information based on the historical operation and maintenance query information, the historical operation and maintenance response information, and the updated text data, and update each second model based on the key feature information, thereby updating the first model.

[0024] In a second aspect, the embodiments of the present disclosure provide a station-city complex operation and maintenance business intelligent retrieval system based on a large model, comprising:

[0025] A data acquisition module configured to acquire operation and maintenance query information related to station-city complex operation and maintenance input by a user;

[0026] A retrieval query module configured to input the operation and maintenance query information into an operation and maintenance large model to query and acquire corresponding operation and maintenance response information; wherein the operation and maintenance large model is obtained by training an original large model based on sample operation and maintenance data of a station-city complex; the original large model comprises a first model and a plurality of second models, the sample operation and maintenance data comprises a plurality of different business type text data related to operation and maintenance and annotation data; the total number of the plurality of second models is the same as the total number of the plurality of different business types;

[0027] And wherein the training process of the original large model comprises: iteratively training each second model independently according to the text data and the annotation data of each business type until the performance parameters of each second model meet a preset condition, wherein the text data of each business type corresponds to one second model; inputting the text data of each business type in the sample operation and maintenance data into each trained second model, constructing a feature set according to the output of each second model, and executing integrated training of the first model based on the feature set to deploy the first model after training as the operation and maintenance large model.

[0028] In a third aspect, the embodiments of the present disclosure provide a computer readable storage medium, which stores a computer program. The computer program is executed by a processor to implement the station-city complex operation and maintenance business intelligent retrieval method based on a large model according to any one of the above embodiments.

[0029] In a fourth aspect, the embodiments of the present disclosure provide an electronic device, which comprises:

[0030] a processor; and

[0031] a memory configured to store a computer program;

[0032] The processor is configured to execute the computer program to implement the station-city complex operation and maintenance business intelligent retrieval method based on a large model according to any one of the above embodiments.

[0033] Compared with the prior art, the technical solutions provided by the embodiments of the present disclosure have the following advantages:

[0034] The station-city complex operation and maintenance business intelligent retrieval method and system based on a large model provided by the embodiments of the present disclosure obtain operation and maintenance query information related to station-city complex operation and maintenance input by a user, input the operation and maintenance query information into an operation and maintenance large model to obtain corresponding operation and maintenance response information. The operation and maintenance large model is obtained by training an original large model based on sample operation and maintenance data of a station-city complex. The original large model includes a first model and a plurality of second models. The sample operation and maintenance data includes text data of a plurality of different business types related to operation and maintenance and annotation data. The total number of the plurality of second models is the same as the total number of the plurality of different business types. The training process of the original large model includes: iteratively training each second model independently according to the text data and the annotation data of each business type until the performance parameters of each second model meet a preset condition, wherein the text data of each business type corresponds to a second model; inputting the text data of each business type in the sample operation and maintenance data into each trained second model, constructing a feature set according to the output of each second model, and executing integrated training of the first model based on the feature set to deploy the first model after training as the operation and maintenance large model. The scheme of the present embodiment can be used as an intelligent operation and maintenance business knowledge retrieval system of a station-city complex, helping operation and maintenance personnel to search for various problems and corresponding solutions directly from the system, which helps to improve the efficiency and accuracy of obtaining operation and maintenance response information such as fault diagnosis results, and assists in improving the work efficiency and accuracy of the personnel, so that the operation and maintenance personnel of the station-city integrated three-dimensional network can quickly and accurately find the relevant solutions when facing problems, thereby shortening the fault recovery time and improving the problem processing efficiency. BRIEF DESCRIPTION OF DRAWINGS

[0035] The accompanying drawings, which are incorporated herein and constitute part of the specification, illustrate embodiments consistent with the present disclosure and serve to explain the principles of the present disclosure together with the specification.

[0036] In order to more clearly illustrate the technical solutions in the embodiments of the present disclosure or the prior art, the accompanying drawings required to be used in the embodiments or prior art description will be briefly introduced. Obviously, for those skilled in the art, other drawings can also be obtained based on these drawings without any creative effort.

[0037] Figure 1 The flow chart of the station-city complex operation and maintenance business intelligent retrieval method based on a large model according to the embodiments of the present disclosure;

[0038] Figure 2 The schematic diagram of the training process of the operation and maintenance large model according to the embodiments of the present disclosure;

[0039] Figure 3 The flow chart of the training method of the operation and maintenance large model according to the embodiments of the present disclosure;

[0040] Figure 4 The schematic diagram of the station-city complex operation and maintenance business intelligent retrieval system based on a large model according to the embodiments of the present disclosure;

[0041] Figure 5 The schematic diagram of the electronic device according to the embodiments of the present disclosure. DETAILED DESCRIPTION

[0042] In order to more clearly illustrate the technical solutions in the embodiments of the present disclosure or the prior art, the accompanying drawings required to be used in the embodiments or prior art description will be briefly introduced. Obviously, for those skilled in the art, other drawings can also be obtained based on these drawings without any creative effort.

[0043] In the following description, many specific details are set forth in order to provide a thorough understanding of the present disclosure. However, the present disclosure can be practiced without the specific details. It is apparent that the examples described in the specification are only some of the embodiments of the present disclosure, not all the embodiments.

[0044] It should be understood that, in the following, "at least one" refers to one or more, and "multiple" refers to two or more. "And / or" is used to describe the association relationship of the associated objects, which means that there can be three relationships, for example, "A and / or B" can represent three cases of only A, only B, and A and B existing at the same time, where A and B can be singular or plural. The character " / " generally represents an "or" relationship between the associated objects before and after it. "At least one of the following" or similar expressions means any combination of these items, including any combination of single or multiple items. For example, at least one of a, b or c can represent a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, and c can be single or multiple.

[0045] The large model related to operation and maintenance in the related art lacks field pertinence. The model trained well on general data is not suitable for operation and maintenance business of various station-city complexes. In addition, the performance is highly dependent on the large model used. If these large models fail to cover the complexity of some queries or the knowledge of a specific field, the accuracy of the operation and maintenance query result is insufficient.

[0046] Figure 1 A flow chart of a station-city complex operation and maintenance business intelligent retrieval method based on a large model is provided for an embodiment of the present disclosure. The method can be executed by an electronic device such as a computer or a mobile terminal, and can specifically include the following steps:

[0047] Step S101: Obtain operation and maintenance query information related to station-city complex operation and maintenance input by a user.

[0048] For example, the operation and maintenance query information can be an inquiry text sentence related to operation and maintenance: "How to adjust the air conditioning system in the station in the summer high-temperature weather to maintain the best energy efficiency and comfort."

[0049] Step S102: Input the operation and maintenance query information into an operation and maintenance large model to query and obtain corresponding operation and maintenance response information; wherein the operation and maintenance large model is obtained by training an original large model based on sample operation and maintenance data of a station-city complex.

[0050] For example, the operation and maintenance response information can be an answer result given to the above inquiry text, such as an operation and maintenance treatment measure or a recommended answer text sentence, for example, “it is recommended that the optimal temperature setting of the air conditioner is 26°, and the running time is set to X”. The original large model can include a first model and a plurality of second models, which can be machine learning models such as support vector machines, convolutional neural networks CNN, long short-term memory networks (LSTM, Long Short-Term Memory), Transformer models, etc. The sample operation and maintenance data includes text data and labeled data of a plurality of different business types related to operation and maintenance.

[0051] Specifically, the station-city complex is the integration of transportation and city, combining transportation hubs and city functions, covering transportation, office, commerce, hotel, cultural facilities, etc. Its types can be divided into high-speed rail station complex, subway station complex, etc. Therefore, it involves various business data such as structure (such as building structure, mechanical and electrical equipment, etc.), environment, energy consumption, people flow, and situation. In order to make the operation and maintenance retrieval scheme of the present embodiment more domain-specific and able to answer user questions about various businesses of the station-city complex, it is necessary to prepare operation and maintenance related data of the station-city complex in advance, for example, by using the five operation state indicators of structure, environment, energy consumption, people flow, and situation to provide targeted decision support and fault response guidance for site operation and maintenance. The pre-prepared data can include data related to the five operation state indicators, such as building facility data, mechanical and electrical equipment details, equipment failure and maintenance records, environmental monitoring data, energy consumption, people flow data, and equipment operation specifications, to ensure comprehensive and accurate data.

[0052] After data cleaning, i.e. deleting duplicates, removing outliers, data standardization, etc., the data is classified according to business type, thereby forming a plurality of text data of different business types. Based on the specific operation and maintenance tasks to be implemented, the text data of each business type can be labeled, such as question and answer pairs, faults and corresponding solutions, etc. This uses actual site operation and maintenance data for the following independent training to ensure that the corresponding model can accurately understand and respond to specific problems in site operation.

[0053] Next, the text content can be subjected to natural language processing to remove irrelevant characters such as HTML tags, special symbols, etc., and to divide continuous text strings into individual lexical units, removing common but meaningless words such as “and”, “is”, etc.

[0054] Finally, the text data is converted into a numerical format that can be processed by machine learning algorithms using feature engineering. In other embodiments, for text data of different business types, appropriate storage systems can also be selected according to data access frequency and data size to store the training data corresponding to each second model respectively.

[0055] The total number of the plurality of second models is the same as the total number of the plurality of different business types, that is, each second model has its corresponding business type training data. For example, business type 1 corresponds to model 1, business type 2 corresponds to model 2, and so on. Referring to Figure 2 As shown, the training process of the original large model can include the following steps:

[0056] According to the text data of each business type and the labeled data, each second model is independently iteratively trained until the performance parameter of each second model meets the preset condition, wherein the text data of each business type corresponds to a second model;

[0057] The text data of each business type in the sample operation and maintenance data is input into each trained second model, and a feature set is formed according to the output of each second model. The first model is trained based on the feature set to deploy the first model after training as the operation and maintenance large model.

[0058] For example, the feature set can be a set of feature vectors corresponding to the output of each second model. The performance parameter of each second model meets the preset condition, which can be that the performance parameter of each second model is greater than or equal to a preset value (such as the loss function value of the model is less than or equal to a preset loss value), which can be set according to specific needs. It can be understood that the preset value corresponding to the performance parameter of each second model can be different, for example, model 1 corresponds to preset value 1, and model 2 corresponds to preset value 2, which can be different from preset value 1, because the operation and maintenance tasks implemented by each second model after training are usually different, and the training data is different.

[0059] For each operation and maintenance task implemented by each second model, the following are some examples:

[0060] 1. Environmental regulation problem solving

[0061] Question and answer scenario: How to adjust the air conditioning system in the station to maintain the best energy efficiency and comfort in the summer high-temperature weather.

[0062] Data training: The model can use historical weather data, passenger flow and air conditioning system performance data to suggest the best temperature setting and running time.

[0063] 2. Peak flow management

[0064] Question and answer scenario: How to adjust the entry and exit management during holidays or large-scale activities to avoid congestion and improve safety.

[0065] Data Training: Utilize past holiday foot traffic data, incident records, and current safety protocols to train the model to provide real-time crowd control strategies and emergency evacuation recommendations.

[0066] 3. Energy Consumption Optimization

[0067] Q&A Scenario: Ask how to reduce energy consumption at night without compromising safety and essential operational needs.

[0068] Data Training: Combine historical energy usage data, night-time operational activities, and device energy efficiency standards to have the model provide energy-saving operation recommendations, such as adjusting lighting and non-critical device usage.

[0069] 4. Fault Response and Maintenance

[0070] Q&A Scenario: Elevator stops functioning, and maintenance personnel need to quickly diagnose the issue and access repair guidelines.

[0071] Data Training: Use the elevator's fault history, repair manuals, and successful cases to train the model to provide fault analysis and repair steps.

[0072] 5. Safety Monitoring and Emergency Response

[0073] Q&A Scenario: Suspicious package is found, ask how to perform a safety check and determine if passengers need to be evacuated.

[0074] Data Training: Utilize safety protocols, historical safety event handling results, and relevant laws and regulations to train the model to provide immediate action guidance and communication strategies. The above is only an example and is not limited thereto.

[0075] After training the different base models according to the business types of the data, that is, a plurality of second models and corresponding text data, due to the fact that the data involved in the station city complex is complex and multiple, and the text data of at least part or all of the business types in the multiple different business types have a correlation relationship with each other. Therefore, in the embodiment, the method of ensemble learning is continued to be used for fusion training, that is, the outputs of these base models are input into a meta model, that is, a first model, as a new feature set, and the task of the meta model is to learn how to most effectively combine the feature information from different base models. In the training phase, first, the performance of each base model is evaluated on an independent validation set (based on part of the data obtained by dividing the sample operation and maintenance data), and then the outputs of the trained base models are used to train the meta model. This process can optimize each model through cross-validation to ensure that each model fully utilizes the data while avoiding overfitting. Finally, the trained meta model is deployed to the actual operation and maintenance environment as an operation and maintenance large model.

[0076] The integrated model can provide more accurate and reliable predictions than individual models, helping to improve the efficiency and accuracy of operational response information such as fault diagnosis results. In addition, a model performance monitoring mechanism and regular update process can be established to adapt to changes in the operation and maintenance environment and new data inputs. For details, please refer to the corresponding description below.

[0077] In this embodiment, the first model and the second model are both large language models (LLM, Large Language Model) that can deeply understand and process natural language text, thereby achieving more accurate semantic analysis when processing operational text data. Through the natural language processing (NLP) capabilities of the large model, valuable feature information such as feature vectors can be automatically extracted from various unstructured text data, such as operational fault reports, maintenance logs, operation manuals, and user feedback, for model training. Ultimately, the operational large model trained can improve the accuracy of operational response information such as fault diagnosis results.

[0078] The above scheme of this embodiment can help operational personnel quickly and accurately locate and solve problems, reduce, for example, device fault handling response time, and can achieve automatic classification of problems and accurate recommendation of solutions. Based on the large model, an intelligent operation and maintenance business knowledge retrieval system for station-city complexes can be established, which can centrally store and manage system problems and solutions, enabling station-city integrated three-dimensional network operation and maintenance personnel to quickly and accurately find relevant solutions when facing problems, thereby shortening fault recovery time and improving problem handling efficiency. In addition, it can accumulate and save team knowledge and experience, and promote knowledge sharing among team members.

[0079] In one embodiment, continuing to refer to Figure 2 As shown, the plurality of second models are pre-trained large models, and the training process of the original large model specifically includes:

[0080] Each of the pre-trained large models is fine-tuned independently based on at least part or all of the text data and the labeled data of each business type until the performance parameters of each pre-trained large model meet the preset conditions to stop training;

[0081] Each pre-trained large model is input with text data of each business type, and a feature set is constructed based on the output of each pre-trained large model. The first model is trained based on the feature set, and the training is ended when the performance parameters of the first model meet the preset conditions. The trained first model is deployed as the operational large model.

[0082] Exemplarily, each second model can be a pre-trained large model that is pre-trained, for example, according to the business type of the text data, using a corresponding different base model, i.e., the second model, which is independently trained, and then fine-tuned for text data of different business types such as structure, environment, energy consumption, people flow, and situation, and then combined with the method of ensemble learning to fuse the feature set composed of the output of each pre-trained large model to perform ensemble training of the first model. By fine-tuning the pre-trained model and combining the feature set composed of the output of each pre-trained large model to train the first model, the operation and maintenance large model obtained by training can further improve the accuracy of operation and maintenance response information such as fault diagnosis results, while improving the efficiency of model training and reducing training costs.

[0083] On the basis of any one of the above embodiments, in an embodiment, in combination with reference Figure 3 The training process of each second model can include the following steps, as shown in the above embodiment.

[0084] Step S301: identifying and determining a plurality of entity keywords and relationships between the plurality of entity keywords from the text data of the sample operation and maintenance data, and constructing a knowledge graph of a graph-based text index based on the plurality of entity keywords and the relationships between the plurality of entity keywords;

[0085] Step S302: The retrieval enhancement generation model adopts a double-layer retrieval mechanism to retrieve and extract query keywords in the text data of each business type, queries target keywords similar to the query keywords from the knowledge graph, inputs the query keywords and the target keywords into the second model to obtain generated content, and updates the second model based on the difference between the generated content and the labeled content.

[0086] Exemplarily, entity such as structured data object such as mechanical and electrical equipment and people flow can be extracted from the text data of the sample operation and maintenance data by entity recognition technology, and entity keywords can be words representing attributes such as the name, location, and energy consumption of mechanical and electrical equipment. The relationship between a plurality of entity keywords can be the relationship between objects indicated by different entities, for example, the relationship between mechanical and electrical equipment (such as an elevator) and people flow. Then a knowledge graph of a graph-based text index can be established.

[0087] The double-layer retrieval mechanism, i.e., extracting global keywords and local keywords in the text data input into the model during training as query keywords, can capture complex interdependent relationships from two aspects of high and low layers, and at the same time, similar keywords can be obtained as auxiliary by combining the previously constructed knowledge graph, which are input into the second model to obtain generated content, and then update the second model, so that the operation and maintenance large model obtained by training can further improve the accuracy of operation and maintenance response information such as fault diagnosis results.

[0088] The precision retrieval capability of the prior art using retrieval augmentation generation (RAG) can improve the natural language generation capability, but the traditional RAG method performs poorly when facing complex or uncommon queries and relies on a predefined document set. For example, the existing RAG technology relies on vector similarity and rearrangement algorithms to retrieve the most similar questions and answers, which is not accurate when facing complex queries or new types of queries related to the operation and maintenance of a complex building complex, resulting in the need for further improvement in the accuracy of the output results of the trained operation and maintenance large model. To this end, a new method is used in the present embodiment to improve the text indexing and retrieval process using a graph structure, and a knowledge graph of the graph-based text index is constructed by extracting entities and their relationships from text data by a large model. In addition, the use of a double-layer retrieval mechanism can locally accurately index specific details, while globally widely indexing and retrieving related topics and concepts, so as to ensure that the text generated by the trained operation and maintenance large model is both accurate and informative, and the accuracy of the operation and maintenance response information such as fault diagnosis results can be further improved.

[0089] In one embodiment, the method further comprises displaying a query user interface, the query user interface comprising a search box for receiving user input of the operation and maintenance query information, the operation and maintenance query information comprising at least operation and maintenance related keywords, categories or dates.

[0090] For example, the present embodiment can provide a search and query scheme that is convenient for user operation, and an intelligent search box is provided in a prominent position of the query user interface. A powerful search engine is also provided to allow users to search in detail by keywords, categories or dates to quickly find related fault solutions and best practice guidelines. In some embodiments, the search algorithm can be optimized to ensure that even when the input is incorrect or the keywords are incomplete, the most relevant documents and solution response information can be returned.

[0091] In some embodiments, knowledge content classification can also be implemented: the knowledge base content is classified according to logic and function, such as structural maintenance, environmental monitoring, etc., and each classification is provided with an intuitive icon and a brief description. Users can use a filter to filter the required content according to the content type (such as guide, case study, operation process), relevance or recent update time. Some embodiments can also provide a multi-level navigation menu to allow users to go deep into specific sub-categories, making it easy for users to accurately find knowledge items in specific fields.

[0092] In one embodiment, the query user interface includes a first control such as a virtual button, and the method further includes: in response to a triggering operation such as a mouse click or a touch click on the first control, displaying common questions and their answers about station city complex operation and maintenance, or, obtaining the user's historical query records, and displaying questions and their answers about station city complex operation and maintenance related to the historical query records.

[0093] For example, in this embodiment, a FAQ (Frequently Asked Questions) display can be provided. A FAQ section is provided in the user interface to display common questions and their answers about station city complex operation and maintenance, helping users quickly find solutions to common problems. In some embodiments, based on the user's browsing and search history query records, the FAQ questions and answers related to the history query records can be recommended and displayed to provide answers to the problems that the user may encounter in advance.

[0094] In one embodiment, the query user interface includes a second control such as a virtual button, and the method further includes: in response to a triggering operation such as a mouse click or a touch click on the second control, displaying the user's historical operation and maintenance query information and the historical operation and maintenance response information provided by the operation and maintenance large model; obtaining the user's input filtering information, and filtering target operation and maintenance query information and / or target operation and maintenance response information from the historical operation and maintenance query information and the historical operation and maintenance response information based on the filtering information.

[0095] In this embodiment, the user's query history records can be displayed. An easily accessible interface section is designed in the user interface to display the user's past operation and maintenance query information and the system's responses, for example, a list view is provided, where each entry contains the query time, the query question, and the system's answer. In some embodiments, the user is allowed to search and filter the history records by conditions such as date, question keywords, or answer type, and the user can also be allowed to export their query history records, which can be in PDF or Excel format. An automatic cleaning option can also be provided to allow the user to set the length of time for which the history records are retained.

[0096] In one embodiment, the method can further include the following steps:

[0097] Collecting the historical operation and maintenance query information input by the user and the historical operation and maintenance response information provided by the operation and maintenance large model, and obtaining updated text data in the text data of the plurality of different business types;

[0098] Extracting key feature information based on the historical operation and maintenance query information, the historical operation and maintenance response information, and the updated text data, updating each second model based on the key feature information, and thereby updating the first model.

[0099] For example, the text data of multiple different business types can be, for example, operation manuals, performance parameters of new equipment, and the like. The updated text data, i.e., the updated operation manuals, performance parameters of new equipment, and the like. In this embodiment, an incremental learning strategy can be used. First, the operation and maintenance query data in the user use process and the historical operation and maintenance response information given by the large model, i.e., the feedback record, are collected. Combined with the updated operation manuals, performance parameters of new equipment, and the like, these new text data are classified according to different business types, and then the model is updated and trained accordingly. The key features such as feature vectors in the classified new text data are extracted, and then the second model and the first model are updated using the key features. The update training process can refer to the foregoing embodiments. At this time, only the training data is different, and the knowledge base, i.e., the knowledge graph, can be updated according to the new text data and feedback during the update training process. For example, if a new type of fault or a new maintenance strategy is detected, the relevant content and model should be updated in time to reflect the latest knowledge and user demand, and measures should be taken to avoid forgetting old knowledge, such as retaining old knowledge through experience replay. Finally, the performance of the model can be evaluated regularly to ensure that the new knowledge is correctly integrated into the training and updating of the model, and the model parameters are updated to improve the overall quality of the operation and maintenance knowledge retrieval system and the accuracy of the response results, i.e., the operation and maintenance response information.

[0100] The above-mentioned scheme in this embodiment can also bring the following beneficial effects:

[0101] (1) Improving problem solving efficiency and accuracy

[0102] The problem solving scheme update training model based on accumulation and update can help operation and maintenance personnel quickly locate and solve problems, reduce fault response time, realize automatic classification of problems and accurate recommendation of solutions, and improve problem solving efficiency and accuracy.

[0103] (2) Improving decision quality

[0104] The data in the knowledge retrieval system can support data analysis and reporting, providing decision support for management. Through analysis of historical data, potential problems and trends can be predicted, and countermeasures can be developed in advance.

[0105] (3) Reducing operation and maintenance costs

[0106] By improving problem solving efficiency and accuracy and reducing the repetitive labor of manpower, the human cost of operation and maintenance can be reduced, and the cost of business interruption and recovery due to improper problem solving can be reduced.

[0107] It should be noted that although the various steps of the methods of the present disclosure are described in a particular order in the accompanying drawings, this does not require or imply that the steps must be performed in this particular order, or that all of the steps shown must be performed to achieve the desired result. In addition or alternatively, certain steps can be omitted, multiple steps can be combined into one step, and / or one step can be divided into multiple steps, etc. In addition, it is also easy to understand that these steps can be executed synchronously or asynchronously, for example, in multiple modules / processes / threads.

[0108] As shown in Figure 4 The embodiment of the present disclosure provides a station-city complex operation and maintenance business intelligent retrieval system based on a large model, which comprises:

[0109] The data acquisition module 401 is configured to acquire operation and maintenance query information related to station-city complex operation and maintenance input by a user;

[0110] The retrieval query module 402 is configured to input the operation and maintenance query information into an operation and maintenance large model to query and acquire corresponding operation and maintenance response information; wherein the operation and maintenance large model is obtained by training an original large model based on sample operation and maintenance data of a station-city complex; the original large model comprises a first model and a plurality of second models, the sample operation and maintenance data comprises text data and annotation data of a plurality of different business types related to operation and maintenance; the total number of the plurality of second models is the same as the total number of the plurality of different business types;

[0111] And wherein the training process of the original large model comprises: iteratively training each second model independently according to the text data and the annotation data of each business type until the performance parameters of each second model meet the preset conditions to stop training, wherein the text data of each business type corresponds to a second model; inputting the text data of each business type in the sample operation and maintenance data into each trained second model, constructing a feature set according to the output of each second model, and performing integrated training of the first model based on the feature set to deploy the first model after training as the operation and maintenance large model.

[0112] In one embodiment, at least part or all of the text data of the plurality of different business types have an association relationship with each other.

[0113] In one embodiment, the plurality of second models are pre-trained large models, and the training process of the original large model specifically comprises:

[0114] Fine-tuning each pre-trained large model independently according to at least part or all of the text data and the annotation data of each business type until the performance parameters of each pre-trained large model meet the preset conditions to stop training.

[0115] The text data of each service type is input into each pre-trained large model, a feature set is constituted according to the output of each pre-trained large model, the first model is trained based on the feature set, and the training is ended when a performance parameter of the first model meets a preset condition; and the first model after the training is ended is deployed as the operation and maintenance large model.

[0116] In one embodiment, the training process of each second model includes:

[0117] A plurality of entity keywords and relationships between the plurality of entity keywords are identified from the text data of the sample operation and maintenance data, and a knowledge graph of a graph-based text index is constructed based on the plurality of entity keywords and the relationships between the plurality of entity keywords.

[0118] The retrieval enhancement generation model adopts a double-layer retrieval mechanism to retrieve and extract query keywords in the text data of each service type, queries target keywords similar to the query keywords from the knowledge graph, inputs the query keywords and the target keywords into the second model to obtain generated content, and updates the second model based on a difference between the generated content and labeled content.

[0119] In one embodiment, the system further includes a display module configured to display a query user interface, the query user interface including a search box configured to receive user input of the operation and maintenance query information, the operation and maintenance query information including at least operation and maintenance related keywords, categories, or dates.

[0120] In one embodiment, the query user interface includes a first control, and the display module of the system is further configured to, in response to a triggering operation on the first control, display common questions and answers about operation and maintenance of a station-city complex, or obtain historical query records of a user and display questions and answers about operation and maintenance of the station-city complex related to the historical query records.

[0121] In one embodiment, the query user interface includes a second control, and the system further includes a filtering module configured to, in response to a triggering operation on the second control, display historical operation and maintenance query information of a user and historical operation and maintenance response information given by the operation and maintenance large model, obtain filtering information input by the user, and filter target operation and maintenance query information and / or target operation and maintenance response information from the historical operation and maintenance query information and the historical operation and maintenance response information based on the filtering information.

[0122] In an embodiment, the system can further comprise an updating module configured to: collect historical operation and maintenance query information input by a user and historical operation and maintenance response information given by the operation and maintenance large model, and acquire updated text data in the text data of the plurality of different business types; extract key feature information based on the historical operation and maintenance query information, the historical operation and maintenance response information, and the updated text data, and update each second model based on the key feature information, so as to update the first model.

[0123] As to the system in the above embodiments, the specific manners in which the respective modules perform operations and the corresponding technical effects brought by the same have been described in detail in the embodiments of the method, and thus will not be described in detail here.

[0124] It should be noted that, although several modules or units of the device for action execution are mentioned in the above detailed description, such division is not mandatory. In fact, according to the embodiments of the present disclosure, the features and functions of two or more modules or units described above can be embodied in one module or unit. Conversely, the features and functions of one module or unit described above can be further divided into a plurality of modules or units for embodiment. The components shown as modules or units can or can not be physical units, i.e., can be located in one place, or can be distributed on multiple network units. Some or all of the modules can be selected to achieve the purpose of the present disclosure according to actual needs. Those of ordinary skill in the art can understand and implement without creative labor.

[0125] The embodiments of the present disclosure also provide a computer readable storage medium having a computer program stored thereon, the computer program being executed by a processor to implement the large model-based station city complex operation and maintenance business intelligent retrieval method of any one of the above embodiments.

[0126] By way of example, the readable storage medium can be, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or apparatus, or any suitable combination of the above. More specific examples (a non-exhaustive list) of the readable storage medium include an electrical connection having one or more wires, a portable disc, 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 disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above.

[0127] The computer readable storage medium can include a data signal transported over a carrier wave and can be baseband or propagated along with carriers. The program code embodied on the computer readable storage medium can be transmitted using any appropriate medium, including but not limited to wireless, wired, optical fiber cable, RF, and the like, or any suitable combination of the foregoing.

[0128] The embodiments of the present disclosure further provide an electronic device, including a processor and a memory, the memory being configured to store a computer program. The processor is configured to execute the computer program to perform the method for intelligent retrieval of station complex operation and maintenance services based on a large model according to any of the above embodiments.

[0129] The electronic device 600 according to the embodiments of the present disclosure will be described below with reference to Figure 5 Figure 5 The electronic device 600 shown is merely an example, and should not impose any limitation on the functions and use range of the embodiments of the present disclosure.

[0130] As shown in Figure 5 , the electronic device 600 is in the form of a general computing device. The components of the electronic device 600 can include, but are not limited to, at least one processing unit 610, at least one storage unit 620, a bus 630 connecting different system components (including the storage unit 620 and the processing unit 610), a display unit 640, and the like.

[0131] The storage unit stores program code that can be executed by the processing unit 610, so that the processing unit 610 performs the steps according to various exemplary embodiments of the present disclosure described in the method embodiment part of the present specification. For example, the processing unit 610 can perform the steps of the method as shown in Figure 1

[0132] The storage unit 620 can include a readable medium in the form of a volatile storage unit, such as a random access memory (RAM) 6201 and / or a cache memory unit 6202, and can further include a read-only memory (ROM) 6203.

[0133] ​​The storage unit 620 can also include a program / utility 6204 having a set of program modules 6205, including but not limited to, an operating system, one or more application programs, other program modules, and program data, each of which can include implementations of the network environment in some embodiments or a combination thereof.

[0134] The bus 630 can represent one or more of several types of bus structures, including a storage bus or bus controller, a peripheral bus, a graphics acceleration port, a processor or local bus using any of a variety of bus architectures.

[0135] The electronic device 600 can also communicate with one or more external devices 700 such as a keyboard or pointing device, a Bluetooth device, etc.; other devices such as printers, scanners, etc.; and / or various types of networks. This communication can occur via the I / O interface 650. Still yet, the electronic device 600 can communicate to one or more devices that enable additional features, such as for example, wireless and / or wired communication features, etc. This communication can occur via the network adapter 660. It should be appreciated that for purposes of this application, a computing "device" can generally refer to any type of electronic device that includes high speed data storage that is similar, in some respects, to a computer system. The bus 630 can be used to communicate communications, data, or control signals between or among the various components of the electronic device 600. The bus 630 can be implemented in any suitable manner, including as a proprietary bus, a PCI bus, a PCI-Express bus, a HyperTransport bus, a USB, etc.

[0136] Those skilled in the art will readily appreciate that the example embodiments described herein can be implemented by software and / or firmware in addition to or instead of hardware. Thus, the technical solutions according to the embodiments of the present disclosure can be embodied in a software product including a number of instructions stored in a nonvolatile storage medium (such as a CD-ROM, a USB, a mobile hard disk, etc.) or a network, and the instructions can be executed by a computing device (such as a personal computer, a server, or a network device, etc.) to enable the computing device to perform the methods according to the embodiments of the present disclosure.

[0137] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0138] The above description is merely a specific embodiment of this disclosure, enabling those skilled in the art to understand or implement it. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this disclosure. Therefore, this disclosure is not to be limited to the embodiments described herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A large model-based station-city complex operation and maintenance business intelligent retrieval method, characterized in that, The method comprises: obtaining user inputted operation and maintenance query information related to station city complex operation and maintenance; inputting the operation and maintenance query information into an operation and maintenance large model to query and obtain corresponding operation and maintenance response information; wherein the operation and maintenance large model is obtained by training an original large model based on sample operation and maintenance data of the station city complex; the original large model comprises a first model and a plurality of second models, the sample operation and maintenance data comprises a plurality of different business type text data and annotation data related to operation and maintenance; the total number of the plurality of second models is the same as the total number of the plurality of different business types; at least part or all of the text data of the plurality of different business types have an association relationship with each other; and wherein the training process of the original large model comprises: iteratively training each second model independently according to the text data and annotation data of each business type until the performance parameters of each second model meet the preset conditions; wherein the text data of each business type corresponds to a second model; inputting the text data of each business type in the sample operation and maintenance data into each trained second model, constructing a feature set according to the output of each second model, and executing integrated training of the first model based on the feature set to deploy the first model after training as the operation and maintenance large model; the training process of each second model comprises: identifying and determining a plurality of entity keywords and the relationship between the plurality of entity keywords from the text data of the sample operation and maintenance data, and constructing a knowledge graph based on the plurality of entity keywords and the relationship between the plurality of entity keywords. The retrieval enhancement generation model adopts a double-layer retrieval mechanism to retrieve and extract query keywords in the text data of each business type, queries target keywords similar to the query keywords from the knowledge graph, inputs the query keywords and the target keywords into the second model to obtain generated content, and updates the second model based on the difference between the generated content and the annotation content.

2. The method of claim 1, wherein, The plurality of second models are pre-trained large models, and the training process of the original large model comprises: fine-tuning each pre-trained large model independently according to at least part or all of the text data and annotation data of each business type until the performance parameters of each pre-trained large model meet the preset conditions; inputting the text data of each business type into each trained pre-trained large model, constructing a feature set according to the output of each pre-trained large model, executing integrated training of the first model based on the feature set, and ending the training when the performance parameters of the first model meet the preset conditions, and deploying the first model after training as the operation and maintenance large model.

3. The method according to claim 1 or 2, characterized in that, The method further comprises: displaying a query user interface, the query user interface comprising a search box for receiving user inputted operation and maintenance query information, the operation and maintenance query information comprising at least operation and maintenance related keywords, classification or date.

4. The method of claim 3, wherein, The query user interface comprises a first control, and the method further comprises: In response to a triggering operation on the first control, common questions and their answer information about station city complex operation and maintenance are displayed, or historical query records of the user are obtained, and questions and their answer information about station city complex operation and maintenance related to the historical query records are displayed.

5. The method of claim 3, wherein, The query user interface includes a second control, and the method further includes: In response to a triggering operation on the second control, historical operation and maintenance query information of the user and historical operation and maintenance response information given by the operation and maintenance large model are displayed; Filtering information input by the user is obtained, and target operation and maintenance query information and / or target operation and maintenance response information are filtered from the historical operation and maintenance query information and the historical operation and maintenance response information based on the filtering information.

6. The method of claim 3, wherein, The method further includes: Historical operation and maintenance query information input by the user and historical operation and maintenance response information given by the operation and maintenance large model are collected, and updated text data in the plurality of text data of different business types are obtained; Key feature information is extracted based on the historical operation and maintenance query information, the historical operation and maintenance response information, and the updated text data, and each second model is updated based on the key feature information, so that the first model is updated.

7. A large model-based station-city complex operation and maintenance business intelligent retrieval system, characterized in that, It includes: A data acquisition module is configured to acquire operation and maintenance query information input by a user and related to station city complex operation and maintenance. A search query module is configured to input the operation and maintenance query information into an operation and maintenance large model to query and acquire corresponding operation and maintenance response information; wherein the operation and maintenance large model is obtained by training an original large model based on sample operation and maintenance data of a station city complex; the original large model includes a first model and a plurality of second models, and the sample operation and maintenance data includes a plurality of text data of different business types related to operation and maintenance and annotation data; the total number of the plurality of second models is the same as the total number of the plurality of different business types; at least part or all of the text data of different business types in the plurality of text data of different business types have an association relationship with each other; and wherein the training process of the original large model includes: iteratively training each second model independently according to the text data of each business type and the annotation data, until the performance parameters of each second model meet a preset condition to stop training, wherein the text data of each business type corresponds to a second model; inputting the text data of each business type in the sample operation and maintenance data into each trained second model, constructing a feature set according to the output of each second model, and executing integrated training of the first model based on the feature set to deploy the first model after training as the operation and maintenance large model; The training process of each second model includes: A plurality of entity keywords and relationships between the plurality of entity keywords are identified and determined from text data of sample operation and maintenance data, and a knowledge graph of a graph-based text index is constructed based on the plurality of entity keywords and the relationships between the plurality of entity keywords; The retrieval enhancement generation model adopts a double-layer retrieval mechanism to retrieve and extract query keywords in text data of each business type, query target keywords similar to the query keywords from the knowledge graph, input the query keywords and the target keywords into the second model to obtain generated content, and update the second model based on differences between the generated content and labeled content.

8. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by a processor to implement the station-city complex operation and maintenance business intelligent retrieval method based on a large model according to any one of claims 1-6.

9. An electronic device, comprising: Comprise: a processor; and a memory for storing a computer program; wherein the processor is configured to execute the computer program to implement the station-city complex operation and maintenance business intelligent retrieval method based on a large model according to any one of claims 1-6.

Images