Train operation risk identification method, computing device and storage medium
By obtaining real-time train monitoring and control data and scheduling operation log data, and using large language models to identify train operation risks, the challenges of traditional systems in risk perception and automation processing are solved, and efficient and intelligent train scheduling safety risk identification and processing are achieved. .
Patent Information
- Application Number
- CN202411209475.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-30
- Publication Date
- 2024-11-29
- Estimated Expiration
- 2044-08-30
AI Technical Summary
The traditional train scheduling risk prevention and control system has challenges in overall risk perception, real-time analysis of unstructured data, operator rationality assessment and automatic generation of risk treatment solutions, especially in stations or CTC systems that do not have centralized control systems. In the extremely station control mode, the manual operation is highly dependent, which can easily lead to safety risks.
By obtaining real-time train monitoring and control data and dispatching operation log data, generating prompt instructions and sending them to the train operation risk detection large language model, obtaining risk identification results, realizing intelligent risk identification and automated processing of scheduling behavior, reducing manual intervention, Improve efficiency.
It realizes the safety risk identification of train dispatch operations with high intelligence and strong reliability, and can detect hidden risks in dispatching behavior in real time, improves train operation safety, and reduces the risk of safety accidents caused by manual operations.
Smart Images

Figure CN119037502A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of train intelligent dispatching and control, and in particular to a train operation risk identification method, computing equipment and storage medium. Background Art
[0002] The compliance, timeliness and accuracy of train dispatching directly determine the safety of train operation. Incorrect dispatching operations are very likely to cause major train accidents. At present, stations equipped with centralized control systems (CTC) have realized the automation of dispatching operations, which has greatly reduced the possibility of human errors. However, in stations without CTC systems and in the emergency station control mode of CTC systems, train dispatching operations still require manual operation by dispatchers. Summary of the invention
[0003] In view of the above problems, the present invention is proposed to provide a train operation risk identification method, computing device and storage medium that overcome the above problems or at least partially solve the above problems.
[0004] According to one aspect of the present invention, a method for identifying train operation risks is provided, comprising:
[0005] Obtain real-time monitoring and control data of trains;
[0006] Collect train dispatch operation log data;
[0007] Generate prompt instructions based on the train real-time monitoring and control data and the train dispatching operation log data;
[0008] The prompt instruction is sent to a large language model for train operation risk detection, and an output result of the large language model for train operation risk detection is obtained, wherein the output result includes a risk identification result for train scheduling behavior.
[0009] According to another aspect of the present invention, there is provided a computing device, comprising: a processor, a memory, a communication interface and a communication bus, wherein the processor, the memory and the communication interface communicate with each other via the communication bus;
[0010] The memory is used to store at least one executable instruction, and the executable instruction enables the processor to perform the following operations:
[0011] Obtain real-time monitoring and control data of trains;
[0012] Collect train dispatch operation log data;
[0013] Generate prompt instructions based on the train real-time monitoring and control data and the train dispatching operation log data;
[0014] The prompt instruction is sent to a large language model for train operation risk detection, and an output result of the large language model for train operation risk detection is obtained, wherein the output result includes a risk identification result for train scheduling behavior.
[0015] According to another aspect of an embodiment of the present invention, a computer storage medium is provided, in which at least one executable instruction is stored, and the executable instruction enables a processor to perform operations corresponding to the above-mentioned train operation risk identification method.
[0016] According to the technical solution provided by the present invention, by acquiring real-time monitoring and control data of the train, collecting train dispatching operation log data, and generating prompt instructions based on the real-time monitoring and control data of the train and the train dispatching operation log data, the prompt instructions are finally sent to the train operation risk detection large language model to obtain the output result of the train operation risk detection large language model, and the output result includes the risk identification result of the train dispatching behavior. It can be seen that the present invention provides a safety risk identification technology for train dispatching operations with high intelligence and strong reliability, which can detect risks hidden in dispatching behaviors in real time, and is of great value in preventing the occurrence of train operation safety problems.
[0017] The above description is only an overview of the technical solution of the present invention. In order to more clearly understand the technical means of the present invention, it can be implemented according to the contents of the specification. In order to make the above and other purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are listed below. BRIEF DESCRIPTION OF THE DRAWINGS
[0018] Various other advantages and benefits will become apparent to those of ordinary skill in the art by reading the detailed description of the preferred embodiments below. The accompanying drawings are only for the purpose of illustrating the preferred embodiments and are not to be considered as limiting the present invention. Moreover, the same reference symbols are used throughout the accompanying drawings to represent the same components. In the accompanying drawings:
[0019] Figure 1 A flow chart showing a method for identifying train operation risks according to an embodiment of the present invention is shown.
[0020] Figure 2 A flow chart showing a method for identifying train operation risks according to the second embodiment of the present invention is shown.
[0021] Figure 3 A flow chart showing a method for identifying train operation risks according to the third embodiment of the present invention is shown.
[0022] Figure 4 A flowchart showing a method for identifying train operation risks according to a fourth embodiment of the present invention is shown.
[0023] Figure 5 A schematic diagram of the structure of a computing device according to a sixth embodiment of the present invention is shown. DETAILED DESCRIPTION
[0024] The exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.
[0025] It should be noted that the operation safety of trains is a key link in railway safety prevention and control. Train operation safety requires consideration of many information dimensions, a wide range of areas, a long time span, and is an extremely complex thinking and decision-making process, which requires very high capabilities and qualities of managers. Although rule-based decision support systems can quickly identify and process risks represented by most structured data, they effectively ensure the safety of train operations. However, traditional risk prevention and control systems still face many challenges in terms of global risk perception, real-time analysis of unstructured data such as voice and text, real-time evaluation of the rationality of operators' operations, and automatic generation of risk disposal plans.
[0026] Embodiment 1
[0027] Figure 1 A flow chart of a train operation risk identification method according to the first embodiment of the present invention is shown. Figure 1 As shown, the method comprises the following steps:
[0028] Step S101: Acquire real-time train monitoring and control data.
[0029] In this embodiment, the real-time monitoring and control data of trains is a combination of multiple data, including train control system data, train reception and dispatching plan data, risk monitoring data, etc. Among them, the train control system data refers to the data generated by the train operation control system (Train Control System), which is mainly responsible for monitoring the running status of the train and ensuring the safe operation of the train. The train control system data usually includes the train position, speed, acceleration, communication information between the train and the track, signal system status, etc. These data can be obtained through on-board equipment (such as transponders, sensors, GPS, etc.) and ground equipment (such as track circuits, wireless communication networks) on the train; the train reception and dispatching plan data includes the departure and arrival schedules of the trains, dispatching instructions, train marshaling information, shunting operation plans, etc. These data are usually formulated by the dispatchers of the dispatching center according to the real-time situation and the preset dispatching rules, and are sent to the relevant stations and trains through the CTC (Centralized Traffic Control) centralized train dispatching system; and the risk monitoring data involves the monitoring of various risk factors that may occur during the operation of the train, including the monitoring of risk factors such as the technical status of the train, the line conditions, natural disasters, and human errors; and the risk monitoring data can be obtained in a variety of ways, such as through the on-board monitoring system, the monitoring equipment beside the track, the meteorological information collection system, etc. The above data can be obtained from multiple information sources through pre-set software and hardware interfaces. The specific acquisition method belongs to the well-known technology in the field and will not be repeated here.
[0030] Real-time acquisition of train control system data, train reception and dispatching plan data, and risk monitoring data is crucial for the artificial intelligence model involved in the present invention to identify train scheduling risks. These data enable the model to update inputs in real time, dynamically adjust predictions, improve accuracy, prevent potential risks, and optimize scheduling decisions.
[0031] Step S102: Collect train dispatching operation log data.
[0032] In this embodiment, an agent can be set to obtain the train dispatching operation log data. The agent can be implemented as a software program or system, which can perform specific tasks, such as data collection, monitoring, analysis, etc. The specific implementation method is not specifically limited in the present invention. The train dispatching log is a detailed document or data record that records various operations and events in the train dispatching process. Exemplary, it generally includes train operation information, dispatching decision information, equipment status information, fault and abnormality information, maintenance activity information, timestamp information, etc.
[0033] According to the historical data provided by the train dispatching operation log data, the model can be helped to deeply understand the historical dispatching patterns that have occurred and enhance its risk prediction ability. In particular, according to step S101 and step S102, the real-time monitoring and control data of the train and the train dispatching operation log data can be obtained at the same time, so that the artificial intelligence large model of the present invention brings the advantages of comprehensiveness, real-time and historical combination in the identification of train dispatching risks, which enhances the model's ability to analyze dispatching patterns and trends, optimizes decision support, speeds up anomaly detection, integrates time series and multi-dimensional data, and can further enhance the model's advanced reasoning and early warning capabilities, providing a more reliable error prevention mechanism for train dispatching.
[0034] Step S103: Generate prompt instructions based on the train real-time monitoring and control data and the train dispatching operation log data.
[0035] It should be noted that the sending frequency of the train dispatching operation log data generation prompt instruction can be diversified and can be adjusted according to system resource configuration or user needs. Specifically, the setting of the frequency needs to consider the system's ability to handle large amounts of data and the processing speed of large language models. Too high a frequency may cause the system to be overloaded, while too low a frequency may cause a delay in risk identification. In some embodiments, the system can dynamically adjust the sending frequency based on the risk assessment results and the current operating status, such as increasing the sending frequency during high-risk periods or when an abnormality occurs, and reducing the sending frequency during low-risk periods.
[0036] In this embodiment, the real-time monitoring and control data of the train and the train dispatching operation log data obtained and used to generate prompt instructions can be either plain text or include various information expressions such as audio, voice, and images, based on the data type that the artificial intelligence big model used can accept and parse. Furthermore, the real-time monitoring and control data of the train and the train dispatching operation log data are assembled into the prompt instructions for the artificial intelligence big model, and other information is combined to generate prompt instructions. Exemplarily, a complete prompt instruction usually includes several key parts, such as background information, task description, input data, expected output, and restrictions or rule descriptions, to ensure the clarity of the information and the integrity of the instructions. The following is an example of a complete prompt instruction according to an embodiment of the present application:
[0037] Background information: In order to comprehensively analyze the risks of scheduling operations, you need to conduct risk assessment based on train operation scheduling data.
[0038] Task Description: Please carefully analyze the scheduling job log information submitted to you below to determine whether there are potential risks in the scheduling job.
[0039] Input data:
[0040] Basic information of the train (e.g. train number, type, starting and ending stations, scheduled departure and arrival times, etc.);
[0041] Train dispatch plan information (including route, stopovers, scheduled timetable, etc.);
[0042] Dispatcher operation log information (records of specific operations, such as dispatch commands, time stamps, abnormal event processing, etc.);
[0043] Information perceived by the risk perception system (such as weather changes, equipment status, operational abnormalities, and other risk indicator information).
[0044] Expected Output: Your output should be in two parts:
[0045] First determine whether there is risk, starting with "risky" or "no risk";
[0046] Based on the input information, an analysis of the reasons why risks exist (or do not exist) is given.
[0047] Limitations or rules: Make sure your analysis is comprehensive, covers all input information, and ensures that your reasoning is accurate.
[0048] Sample output:
[0049] There are risks.
[0050] Cause analysis: According to the dispatcher's operation log information, it was found that the stop time of train J123 at XX station exceeded the planned schedule by 7 minutes, and the risk perception system also reported that the station section was affected by bad weather recently, resulting in unstable equipment operation. Combined with the train dispatch plan information, it can be seen that the train is close to the subsequent train and there is a possibility of further delay, so it is determined to have an operation risk.
[0051] Please generate analysis results according to the above prompts.
[0052] From this example, it can be seen that in the prompt instruction, the obtained or collected real-time monitoring and control data of the train and the train scheduling operation log data are integrated into the input data to serve as the basis for the artificial intelligence large model to analyze and judge risks.
[0053] Step S104: Send the prompt instruction to the large language model for train operation risk detection, and obtain the output result of the large language model for train operation risk detection, wherein the output result includes the risk identification result of the train scheduling behavior.
[0054] In this embodiment, the prompt instruction generated in step S103 is sent to the artificial intelligence big model, and the artificial intelligence big model will automatically parse and process the content in the prompt instruction, and finally give its output result on whether there are hidden dangers or risks in the current train scheduling behavior. Exemplarily, the artificial intelligence big model used in this embodiment can be, for example, Llama, ChatGLM, ERNIE, GPT series big models, etc. These big models have been widely used in this field and can also be suitable for implementing the technical solution of this embodiment. Regarding the working mechanism and architecture of each big model itself, this embodiment does not make too many restrictions; in addition, as the desired output result form of the big model included in the prompt instruction in step S103, the output result of the big language model in this embodiment can be, for example, the following examples:
[0055] Risk conclusion: There is risk.
[0056] Cause analysis: According to the XXX operation log information, it was found that the train No. XX stayed at the XX station for more than XX minutes in the planned schedule, and the risk perception system also reported that the station section was recently affected by bad weather, resulting in unstable equipment operation. Combined with the train dispatch plan information, it can be seen that the train is close to the subsequent train and there is a possibility of further delay, so it is determined to have XX risk.
[0057] The train operation risk identification method of this embodiment integrates real-time monitoring data and historical operation log data in the prompt instructions sent to the large language model, which not only provides comprehensive and multi-dimensional train scheduling information, but also integrates time series and multi-dimensional data, significantly improving the accuracy and early warning capabilities of risk prediction, enabling the model to deeply understand the scheduling mode, and optimize decision support and abnormal detection speed. This enables automated processing of train operation risk identification, reduces manual intervention, and improves efficiency; in addition, the clear output format also facilitates quick understanding and decision-making.
[0058] Embodiment 2
[0059] Figure 2 A flow chart of a train operation risk identification method according to Embodiment 2 of the present invention is shown.
[0060] Embodiment 2 is a further refinement of step S102 in embodiment 1.
[0061] like Figure 2 As shown, the method comprises the following steps:
[0062] Step S201: Setting an agent module, wherein the agent module is configured to collect the train dispatching operation log data; the train dispatching operation log data includes: train dispatching log text data and train dispatching instruction audio data.
[0063] As mentioned in step S102, the train dispatching operation log data can be obtained by setting an agent. Voice data is an important part of the dispatching log data. Especially in some emergencies and very normal situations, dispatching instructions are transmitted by voice, and some matters that require confirmation by the dispatcher and the train driver are also carried out by voice. Therefore, the processing of voice data is very important for the identification of dispatching behavior risks and must not be ignored.
[0064] To this end, further optionally, step S202 is proposed: converting the train dispatching instruction audio data into second train dispatching instruction text data.
[0065] In order to better integrate the text data and audio data in the train dispatching log data so that the artificial intelligence large model can better understand and respond, it is necessary to convert the train dispatching instruction audio data into train dispatching instruction text data. This step can be implemented using a commonly used speech recognition model with TTS (text to speech). No specific limitation is made here. It can be understood that the speech recognition model can also be a pre-trained speech recognition model that conforms to the train dispatching scenario to make the recognition result more accurate.
[0066] Furthermore, step S203 is proposed: the converting of the train dispatching instruction audio data into the second train dispatching instruction text data further includes: after converting the train dispatching instruction audio data into the first train dispatching instruction text data, formatting the first train dispatching instruction text data to obtain the formatted second train dispatching instruction text data.
[0067] The audio data of train dispatching instructions often has a huge amount of data and also contains a lot of useless information. If the text data converted from these audio data is used directly, it will not only waste system resources, but may also seriously affect the accuracy of risk identification by the artificial intelligence large model. In view of this, it is also necessary to format the first train instruction text data obtained after conversion to obtain the formatted second train dispatching instruction text data. In other words, it is necessary to summarize and refine the first train instruction text data after audio conversion, extract and integrate useful information, and eliminate useless information.
[0068] In order to implement the above steps, there should also be step S204: the converting of the train dispatching instruction audio data into the second train dispatching instruction text data further includes:
[0069] The train dispatching instruction audio data is converted into first train dispatching instruction text data using a speech recognition model, and semantic extraction of the first train dispatching instruction text data is performed using a semantic extraction large language model to obtain formatted second train dispatching instruction text data. In some embodiments, the formatting operation includes grammar checking, text data format unification, text data structuring, data integrity verification, etc., which ensures the standardization and consistency of the train dispatching instruction text data and improves the efficiency and accuracy of data processing.
[0070] Specifically, in the audio data processing flow of train dispatching instructions, the data can be processed using a speech recognition model and a semantic extraction large language model respectively. Exemplarily, the speech recognition model can be implemented using a common TTS model or tool, such as Google Cloud Speech-to-Text, Amazon Transcribe, Baidu DeepSpeech, etc., and the semantic extraction large language model can be, for example, BERT, ERNIE, GPT, etc., which are not specifically limited here.
[0071] In Example 2, by converting the voice data in the dispatch log into text and using a large language model (LLM) for semantic extraction, firstly, it is ensured that in emergency and abnormal situations, the key dispatch instructions transmitted by voice and the confirmation matters between the train drivers can be accurately captured. Secondly, the speech recognition model converts these voice data into analyzable text form, laying the foundation for further semantic understanding and risk assessment. Subsequently, the LLM conducts in-depth analysis of the text, extracts key information, and realizes formatted output of the data, which not only improves the readability and operability of the dispatch instructions, but also enhances the accuracy and response speed of risk identification, and improves the utilization efficiency of the dispatch log data and the quality of decision support.
[0072] Embodiment 3
[0073] Figure 3 A flow chart of a train operation risk identification method according to Embodiment 3 of the present invention is shown.
[0074] Embodiment 3 is a further refinement of step S103 in embodiment 1.
[0075] like Figure 3 As shown, the method comprises the following steps:
[0076] Step S301: Generating prompt instructions based on the train real-time monitoring and control data and the train dispatching operation log data includes:
[0077] The train real-time monitoring and control data and the train scheduling operation log data are configured as at least part of the input data in the prompt instruction, and are used as at least part of the data analysis basis for indicating the output of the execution result of the large language model for train operation risk detection.
[0078] The real-time train monitoring and control data and the train dispatching operation log data are integrated and assembled into prompt instructions as part of the input data in the prompt instructions. It is necessary to clearly inform the artificial intelligence big model in the prompt instructions that it should perform reasoning and analysis based on these input data when making train operation risk identification and judgment, so as to be used as at least part of the data analysis basis for indicating the output of the execution result of the train operation risk detection big language model.
[0079] Optionally, the method further includes step S302: generating a prompt instruction based on the train real-time monitoring and control data and the train dispatching operation log data further includes:
[0080] The train real-time monitoring and control data and the train dispatching operation log data are configured to be used to retrieve and obtain text data matching the train real-time monitoring and control data and the train dispatching operation log data from a train dispatching related knowledge database, and the text data matching the train real-time monitoring and control data and the train dispatching operation log data are configured as at least part of the input data in the prompt instruction, and are used as at least part of the data analysis basis for indicating the output of the execution result of the large language model for train operation risk detection.
[0081] In this step, the obtained or collected real-time monitoring and control data of the train and the train dispatching operation log data are used to enter the pre-set train dispatching related knowledge database for retrieval to obtain text data matching the above data, and these text data are also assembled into the prompt instruction as another part of the input data in the prompt instruction, and it is necessary to clearly inform the artificial intelligence large model in the prompt instruction that it should be based on these input data for reasoning and analysis in the train operation risk identification and judgment, so as to be used as at least part of the data analysis basis for indicating the output of the execution result of the large language model of the train operation risk detection. The train dispatching related knowledge database can be, for example, a specially constructed database that can provide general train dispatching related knowledge, or it can be an Internet database, etc., which is not specifically limited in this step.
[0082] By searching the database of train dispatch-related knowledge, the artificial intelligence model can obtain supplementary information that is highly relevant to the current dispatch situation, enhancing its ability to understand complex dispatch scenarios and generalize information processing. In addition, clear prompts can also guide the model on how to use these data for effective reasoning, ensuring the pertinence and accuracy of the analysis. This comprehensive data-driven approach not only improves the accuracy of abnormal behavior identification, but also provides strong support for dispatching decisions, thereby optimizing the safety and efficiency of train operations.
[0083] Optionally, the method further includes step S303: the train real-time monitoring and control data and the train dispatching operation log data are configured to be used for retrieving and obtaining text data matching the train real-time monitoring and control data and the train dispatching operation log data from a train dispatching related knowledge database, including:
[0084] The train scheduling related knowledge database includes an unstructured data vector database and / or a structured data database.
[0085] In this embodiment, the knowledge data stored in the database involved includes structured and unstructured data. Exemplarily, the structured data includes basic information of trains and stations, equipment information, risk identification information, scheduling plan information, etc.; and the unstructured information includes documents such as scheduling operation regulations, scheduling responsibilities, and risk management manuals. Optionally, the unstructured data vector database and the structured data database can be constructed in the following ways, including:
[0086] Data preparation: distinguishing between structured data required for train dispatching (including basic information of trains and stations, equipment information, risk identification information, etc.) and unstructured data (such as dispatching operation regulations, risk management manuals and other documents and materials);
[0087] Vector database construction: natural language processing of unstructured data, conversion into vector representation, and storage in the vector database to support efficient fuzzy retrieval and similarity analysis;
[0088] Structured database construction, importing structured data into a structured database to facilitate fast and accurate data retrieval functions.
[0089] For this embodiment, integrating the structured database and the vector database can significantly optimize the efficiency and quality of train scheduling risk identification.
[0090] Optionally, the method further comprises step S304: before the train real-time monitoring and control data and the train dispatching operation log data are configured to be used for retrieving and obtaining text data matching the train real-time monitoring and control data and the train dispatching operation log data from a train dispatching related knowledge database, the method further comprises:
[0091] The train real-time monitoring and control data and the train dispatching operation log data are text-vectorized.
[0092] In order to achieve good retrieval effect and information matching, the methods involved in this step may include:
[0093] Build a knowledge index: vectorize the knowledge from databases, documents, web pages, and knowledge graphs and store them in a vector database. This step is to convert the knowledge into a machine-processable format and store it for subsequent retrieval.
[0094] Retrieval fusion: The train real-time monitoring and control data and the train dispatching operation log data are vectorized, and based on the semantic vector similarity, the document blocks most relevant to the above vector data, such as the first n document blocks, are retrieved from the vector database. The original question and the retrieved document blocks are fused to generate the final retrieval results and output.
[0095] In the embodiment of the present application, the accuracy and reliability of the large model answer are enhanced by using RAG (Retrieval-Augmented Generation) technology. RAG, or Retrieval-Augmented Generation Model, is a method of combining retrieval mechanism and generation model in the field of natural language processing (NLP). It aims to assist in generating more accurate and richer text content by retrieving relevant information. Its core steps include knowledge indexing, retrieval fusion, and content generation. Thus, the RAG module is used to generate the rules of the scene and the relevant basic knowledge according to the characteristics of the scene.
[0096] Exemplarily, the knowledge required for train operation safety includes structured and unstructured data. The results output by the retrieval process in this step can be in the following forms: structured data includes basic information of trains and stations, equipment information, risk identification information, scheduling plan information, etc. Unstructured information includes documents such as scheduling operation regulations, scheduling responsibilities, and risk management manuals.
[0097] Therefore, RAG integrates the capabilities of retrieval and generation, and can effectively extract key information from large amounts of data and generate accurate risk assessments, greatly improving the accuracy of risk identification and the real-time update capabilities of the model. It not only improves the level of security assurance, but also enhances the transparency and credibility of decision-making, and can effectively solve the illusion problem of large models.
[0098] In this embodiment, based on the integration of real-time train monitoring and control data and dispatching operation logs, the use of knowledge database retrieval can provide supplementary information that is highly relevant to the dispatching situation, enhancing the model's understanding of complex scenarios. In addition, clear prompt instructions also ensure the accuracy of the analysis, improve the accuracy of identifying abnormal behaviors, and provide strong support for dispatching decisions, optimizing the safety and efficiency of train operations. The method of integrating structured and unstructured data improves the generalization ability of the model, while the automated retrieval and data processing process improves efficiency and reduces human errors. In addition, through text vectorization and knowledge index construction, the efficiency and quality of retrieval are further improved, ensuring that the model can quickly adapt to new dispatching situations and knowledge updates.
[0099] Embodiment 4
[0100] Figure 4 A flow chart of a train operation risk identification method according to a fourth embodiment of the present invention is shown.
[0101] Embodiment 4 is a further refinement of step S104 in embodiment 1.
[0102] like Figure 4 As shown, the method comprises the following steps:
[0103] Step S401: sending the prompt instruction to the large language model for train operation risk detection, obtaining the output result of the large language model for train operation risk detection, the output result including the risk identification result of the train dispatching behavior, including:
[0104] The train operation risk detection large language model is a fine-tuned large language model.
[0105] In this step, by adjusting the parameters of the pre-trained model on the data set in the field of train scheduling, the performance of the model on the specific task of vehicle operation risk identification of the present invention can be improved. This step can refine and adaptively adjust the model and correct errors by optimizing parameters on new data related to the target task.
[0106] In some examples of the embodiments of the present application, a model of train operation risk control based on LLM, Train Security Large Model (TSLM), is provided. Specifically, the safety control of railway operation is divided into different scenarios, and TSLM is applied to perform real-time analysis of each scenario to determine whether there is a risk in the train operation scenario. The edge perception system transmits the monitored data to LLM in real time. TSLM encapsulates the data into a specific scenario set based on the characteristics of the detection data. The scenario set includes the rules and basic knowledge required to process the detection data. Then, TSLM integrates COT (Chain-of-Thought) technology with the scenario data to generate instruction data that can be used for LLM reasoning, and finally infers and outputs the risk identification results and disposal suggestions of the scenario.
[0107] Specifically, a TSLM-backbone that supports TSLM reasoning is fine-tuned based on the open source LLM. On the one hand, TSLM-backbone can solve the problem that general LLM lacks expertise in the vertical field of train operation safety control, and can improve LLM's understanding of domain knowledge. On the other hand, this study uses the COT method to enhance and fine-tune LLM, making LLM's reasoning ability more significant. In addition to effectively enhancing the reasoning effect, the role of TSLM-backbone also provides support for the private deployment of TSLM.
[0108] Here, the scenario-based train operation risk identification method divides the risks of train operation into different scenarios and performs risk prevention and control according to the characteristics of different scenarios. The risk identification framework TSLM implemented by LLM reasoning can identify the risks of train operation scenarios.
[0109] Optionally, the method further includes step S402: the train operation risk detection large language model is a fine-tuned large language model, including:
[0110] Using a few-sample thought chaining prompting method, a training dataset for the fine-tuning is generated.
[0111] Train dispatching is a comprehensive job, and the knowledge required for this job covers a variety of railway business knowledge. In order for LLM to fully understand the knowledge in the dispatching field, it is necessary to apply a variety of railway business knowledge for fine-tuning. In the present invention, a large amount of text content related to railway dispatching is collected for model training, including but not limited to the following for example:
[0112]
[0113]
[0114] In some examples of the embodiments of the present application, for the collection of fine-tuning data, large models such as GPT-4 can be used to generate high-quality, wide-coverage training data to ensure data diversity and annotation accuracy.
[0115] The traditional method of manual construction is very costly. Therefore, the present invention proposes a method of using a large model to construct a fine-tuning instruction data set. This method uses a few-shot chain-of-thought prompting method to allow LLM to generate training data with thought chains under the guidance of prompt words, so that the model can discover problems in scheduling behavior through reasoning. The following is an example of the method of constructing a fine-tuning data set in the present invention:
[0116] In some embodiments, the accuracy of model reasoning is improved by comprehensively adopting methods such as RAG, fine-tuning, COT, etc. The accuracy and reliability of large model answers are enhanced by adopting RAG technology.
[0117] Fine-tuning can make large models perform better in vertical fields. By adjusting the parameters of pre-trained models on data sets in specific fields, the models can master professional terms and knowledge, improve the accuracy and relevance of tasks, enhance problem-solving capabilities, and quickly adapt to field needs, effectively reducing development costs and time. This method can fine-tune the model to refine and adapt it, and correct errors by optimizing parameters on new data related to the target task.
[0118] Through COT technology, LLM's reasoning ability and complex problem-solving ability can be improved. By breaking down the problem step by step, COT guides the model to solve the task in logical steps, thereby improving the accuracy and consistency of the answer. In addition, COT also enhances the model's performance in multi-step tasks, making it more effective in handling problems that require reasoning and multi-level analysis.
[0119] Therefore, the mixed use of multiple methods such as RAG, Fine-tuning and COT can significantly improve the application effect of large models. This combination method not only improves the accuracy and reliability of the model in vertical fields, but also enhances its reasoning and problem-solving capabilities, making the model more efficient and flexible when dealing with complex tasks, thereby better meeting various practical needs. Therefore, TSLM divides the risks of train operation into different scenarios for refined prevention and control through scenario-based risk identification methods, and combines RAG, Fine-tuning and COT methods to enhance the model's reasoning and risk identification capabilities.
[0120] First, a scenario set is constructed. All scheduling operations occur in certain scenarios and must strictly follow the rules of the rule base. Scheduling operations that do not comply with the rules will bring potential safety hazards. Based on this consideration, the present invention uses scheduling rules as seed data to generate corresponding fine-tuning instruction data.
[0121] Further, the structure of the training samples is determined. One of the purposes of the present invention is to use fine-tuning data to train a model for risk identification of dispatching behavior, which is used to infer and analyze the dispatcher's dispatch log to identify whether his dispatching behavior poses a safety risk. Therefore, when designing the fine-tuning data set, this special scenario requirement must be fully considered to ensure that the model can effectively learn and predict potential risks. The following is the preferred format:
[0122] The input part corresponds to the scenario set. The input part of the scenario must contain a series of feature sequences related to train scheduling: 1) The rules and basic information required for reasoning, such as basic information of the station, basic information of the train, and scheduling plan. This part of information is the basis for LLM reasoning judgment. 2) Scheduling log. A time-ordered scheduling behavior log. Including global data such as the scheduling operation instruction log obtained from the scheduling system and the monitoring data log provided by the natural disaster prevention system.
[0123] In some examples of the embodiments of the present application, a scenario is a description of the state of the train operating environment. A scenario consists of the interaction of different entities in multiple dimensions. Entities can be described by static attribute data and dynamic behavior data. Therefore, a scenario can be defined as a collection of these elements:
[0124] S={E i (P i , B i (t))|i=1, 2, 3,..., n} (1)
[0125] Where: P i Represents the attribute data of the i-th entity, B i (t) represents the behavior data of the ith entity at time t, E i (P i , B i ) indicates that the entity is represented by the attribute data P i and behavioral data B i The i-th element of , where n is the total number of entities in the scene.
[0126] As can be seen from formula (1), the key to constructing the scenario S(t) is to construct B(t). The data of B(t) can be collected from the log data of the existing control system. Since the data containing risk data in the real data is very sparse, when generating behavioral data, we should consider generating sample data that can trigger risks to train the model's ability to deal with risks. Trains operate according to certain rules. If they operate within the scope allowed by the rules, they are safe. If they exceed the scope allowed by the rules, risks will arise. Therefore, this study uses the train operation rules as seed data to generate the behavioral data of each S(t).
[0127] Furthermore, regarding the construction of real-time scenarios, it can be seen from formula (1) that since the attributes of entities in the scenario are static, this part of the content is easy to collect. The difficulty lies in obtaining the real-time data B(t) of the scenario. B(t) represents the set of behavioral events that may cause risks, which can be expressed as a combination of multiple elements:
[0128] B(t)={A,T,B} (2)
[0129] In the formula, A represents the subject of the behavior, which includes people and equipment. People include train drivers, dispatchers, station attendants, and signalmen, and equipment mainly includes various control systems and edge sensing systems. T represents the time when the behavior occurs. B represents the specific behavior, the action or activity actually performed by the subject of the behavior, which is the core element of the event.
[0130] The following is a prompt word form used in the present invention for a rainfall scene:
[0131] Rules: When there is rain, and the rainfall in key flood control areas reaches 45mm or more in one hour, the train speed limit is 120km / h......
[0132] Task instructions: Please generate the dispatcher's operation log according to the input rules and log data. The dispatcher will need to process each system log. The format of the processed log is [behavior subject], [time], [behavior that occurred].
[0133] To be more specific, here is an example of generating a dispatch action based on rainfall log data:
[0134] [{{"Subject":"Rain monitoring system","Time":"18:10:54","Behavior":"Monitored rainfall in 1 hour: 47.1mm"}},
[0135] …
[0136] ].
[0137] Furthermore, the dispatcher's processing log data is generated as follows: [
[0139] {{"Subject":"Rain monitoring system","Time":"18:10:54","Monitored rainfall in 1 hour: 47.1mm"}},
[0140] {{"Subject":"Train dispatcher","Time":"18:10:54","Action":"Issued train speed limit order, speed limit 45km / h"}},
[0141] …
[0142] ].
[0143] The core of this prompt word is to let LLM simulate the dispatcher's decision-making behavior according to the rules and log data of the monitoring system, and randomly generate some behavioral data that does not meet the specifications.
[0144] As a further optimization of the embodiment of the present application, in the RAG-based scenario construction, RAG is the key link of RSLM, providing institutional regulations and necessary basic knowledge for RSLM reasoning, and is the basis of model reasoning. In order to optimize the performance of the RAG model in train operation risk identification, TSLM uses both structured databases and vector databases. This "dual database" structure has significant advantages in many aspects. First, it enhances the retrieval capability. By storing structured and unstructured data separately, the model can retrieve the required information more effectively and improve the relevance and accuracy of the answers. Secondly, it improves flexibility and adaptability, enabling the model to flexibly handle different types of data, adapt to changing information needs, and facilitate optimization and updating for specific data types. In addition, the combination of the two databases improves the comprehensive utilization of various types of data in train scheduling, providing a richer and more accurate knowledge background for risk identification. Finally, the rich information retrieved from the two databases enables the RAG model to generate more accurate risk assessment results and scheduling recommendations, supporting high-quality decision-making. These advantages together significantly optimize the efficiency and quality of train operation risk identification.
[0145] In some examples of the embodiments of the present application, the risk of a scene can be defined as the potential danger caused by the scene state characteristics reaching a certain threshold. Combined with the previous definition of the scene, the risk of the scene can be defined as follows:
[0146] Assuming the state feature in a specific scene is Fs, the state feature of the scene can be expressed as:
[0147] F S =f(E i (P i , B i (t))), i=1,2,3,...,n (3)
[0148] Where f is the risk identification function of entity E. T is the safety threshold, which is generally determined by business rules. When Fs is within the safety threshold T, scenario S is safe. When Fs exceeds T, risk R is generated. When s Assuming that there are m risk scenarios in the train operation process, the risk R faced by the train operation can be expressed as:
[0149]
[0150] Where: m represents the number of train operation scenarios, T j represents the safety threshold of the jth scenario, F j represents the characteristics of the jth scene. This formula represents all The set of risk states of the conditions, that is, as long as the conditions If it is established, it means that the risk of the jth scenario is triggered, then R j will be included in this risk set R.
[0151] The Target section includes two parts: risk process and conclusion. Risk analysis is strictly based on the materials provided by the scenario description, and the thinking chain is used to perform reasoning analysis to analyze whether the behavior reflected in the log is risky. Finally, a conclusion is drawn as to whether there is a risk.
[0152] More specifically, the risk prevention and control of train operation is divided into two parts: identification and control. Usually these two parts are carried out separately, identification first, and then processing. At present, many studies only focus on risk identification and early warning, and there are fewer studies on risk processing. Based on the TSLM framework proposed in the embodiment of the present application, an "end-to-end" risk prevention and control solution is provided by adopting the COT reasoning mechanism, which can complete the two processes of risk identification and risk control in one reasoning process. TSLM obtains the behavioral data of the scene S in real time, constructs a real-time scene S(t), and combines the rules of the scene and related basic knowledge to form the input of TSLM, and obtains the identification result after basic LLM reasoning. It can be expressed as:
[0153] Re(d, r, c) = F risk (P risk (S(t), R, K)) (5)
[0154] Where: S(t) represents the real-time scene set, R represents the scene rule, K represents the necessary basic knowledge, R and K are generated by the RAG module according to the characteristics of the scene, P risk Indicates the construction method of risk reasoning prompt. riskRe(d, r, c) is the result of the reasoning, where d represents the reasoning process chain of the large model, r represents the risk identification result, and c represents the risk disposal measure. The core of formula (5) is to apply the reasoning ability of LLM in the scenario space defined by the scenario, reason on the multidimensional information provided by the scenario, and output the corresponding reasoning results to maximize the accuracy of risk control.
[0155] Finally, create prompts to generate fine-tuning instruction data. Create a prompt for the training data set of a model for scheduling behavior risk identification based on the Few-shot Chain-of-Thought Prompting method. The instruction part mainly clarifies the generated instructions. The input part includes control rules and S(t). The output part includes three parts: reasoning process, risk identification results, and risk response plan. For example,
[0156] The following shows a prompt word form of the COT fine-tuning data generation Prompt example used in the present invention:
[0157] Task instructions: Please reason and analyze the behavioral events in the scenario based on the following rule scenario knowledge to determine whether there are risks in the system.
[0158] Requirements: Use the thinking chain method, strictly follow the provided rules, first perform logical reasoning, then obtain the identification results, and give suggestions for handling the risk. The output result format is:
[0159] {
[0160] "Risk Reasoning": Reasoning process,
[0161] "Identification result": [risk exists, no risk]
[0162] "Disposal suggestions": If there is a risk, output the risk disposal suggestions. If there is no risk, do not output it.
[0163] }
[0164] Rules and knowledge: In case of rainy weather, when the rainfall in key flood control areas reaches 45mm or more and less than 60mm in one hour, the train speed limit is 120km / h; ...
[0165] Scenario: [
[0167] {"Subject":"Rain Monitoring System","Time":"2024-05-14 18:10:54","Behavior":"The observed rainfall in 1 hour is: 47.1mm"},
[0168] {"Subject":"Train dispatcher","Time":"2024-05-14 18:10:54","Action":"Issued a train speed limit order, speed limit 45km / h"},
[0169] …
[0170] ].
[0171] In the embodiment of the present application, the TSLM framework uses the LLM's chain of reasoning mechanism to simultaneously complete the output of risk identification and risk treatment in one reasoning process. The COT mechanism of LLM can simulate the step-by-step reasoning process of human thinking and draw conclusions through multi-step reasoning. Through this mechanism, TSLM gradually processes the input multi-source information, layer by layer, to achieve a comprehensive analysis of complex problems. In order to guide the large model to gradually reason, identify and process the risks in train operation, combined with P in formula (5) risk Given the definition of (S(t), R, K) input and output Re(d, r, c), this study proposes a template for constructing this PROMPT.
[0172] The following is an example template for a COT-based prompt:
[0173]
[0174] Through the above method, the TSLM framework realizes the reasoning of risk identification and processing in a reasoning process with the help of LLM's thinking chain reasoning mechanism.
[0175] In this step, a large model is used to construct a fine-tuning instruction data set, which provides an efficient and accurate solution for processing large amounts of complex data. It is particularly suitable for application scenarios such as the present invention for train operation risk identification, which require extracting information from rich but unstructured text content and constructing a training data set. Thus, a train operation risk monitoring model is trained using fine-tuning data to realize reasoning about S(t) to identify whether there is a safety risk in the train under the S(t) state.
[0176] By introducing RAG nodes and COT structures, the TSLM model has significantly improved the accuracy of the knowledge involved in the prompt, thereby comprehensively optimizing the analysis effect of TSLM. Specifically, this architectural improvement enables the model to integrate and apply relevant knowledge more accurately, improving the overall performance of risk identification and interpretation.
[0177] Furthermore, the method further includes step S403: the train operation risk detection large language model is a fine-tuned large language model, including:
[0178] The fine tuning is performed using the LORA method.
[0179] An important technical problem to be solved by the present invention is to use fine-tuning technology to fine-tune a large model for train dispatching behavior risk identification. Although various embodiments of the present invention obtain a large amount of fine-tuning data from public materials, these data are still relatively small compared to the amount of data required for large model training.
[0180] By adopting the Low-Rank Adaptation (LoRA) fine-tuning method, while maintaining the original model capabilities, its adaptability to specific tasks is enhanced, the fine-tuning efficiency is improved and overfitting is avoided, thus ensuring the original reasoning capabilities of the basic LLM.
[0181] Lora fine-tuning technology provides an effective means for optimizing large pre-trained models under data constraints. By introducing low-rank updates in specific layers of the model, Lora can achieve efficient parameter adaptation rather than full parameter adjustment, thereby reducing computational costs and improving fine-tuning efficiency. This method is particularly suitable for dealing with computational resource limitations and overfitting risk issues in fine-tuning large models. Due to the complexity of identifying risks in train scheduling behavior and the limitations of data available for fine-tuning, the present invention uses Lora to fine-tune the benchmark large model.
[0182] Optionally, the method of fine-tuning a large model using the Lora method in the present invention includes:
[0183] Fine-tuning data preparation: Prepare the fine-tuning data set based on the business data required for the fine-tuning task. The quality of the fine-tuning data is a decisive factor in the effect of model fine-tuning.
[0184] Model selection: Select a pre-trained benchmark model, such as Llama, ChatGLM, ERNIE, etc., and choose one based on the characteristics of the task that needs to be fine-tuned.
[0185] Fine-tuning training: It mainly involves identifying the adjustment layers, initializing a set of low-rank matrices in the selected layers, and updating the low-rank matrices with the help of the gradient descent optimization algorithm.
[0186] Evaluation and Tuning: Evaluate the model performance on the validation set, monitor the fine-tuning process and avoid overfitting. Make necessary adjustments based on the model's performance on the validation set and repeat the fine-tuning process to optimize performance.
[0187] Lora provides an effective means of optimization under data constraints. By introducing low-rank updates at specific layers of the model, Lora can achieve efficient parameter adaptation rather than full parameter adjustment, thereby reducing computational costs and improving fine-tuning efficiency. This method is particularly suitable for dealing with computational resource limitations and overfitting risk issues in fine-tuning large models. Due to the complexity of train operation risk identification and the limitations of data available for fine-tuning, this study uses Lora to fine-tune the benchmark large model. The following table shows the main parameter configurations for this fine-tuning.
[0188]
[0189] In this embodiment, for the train operation risk identification scenario, through fine-tuning on a specific data set, the model can more accurately understand and adapt to the characteristics and rules of the train scheduling field; and the small sample thinking chain prompt method solves the problem of insufficient labeled data. At the same time, the integration of rich railway scheduling knowledge ensures the model's comprehensive understanding of safety standards; in addition, LORA fine-tuning technology optimizes the model under data constraints, reduces computing costs, reduces overfitting risks, and ensures the generalization ability of the model and the reliability of decision support through continuous evaluation and tuning.
[0190] As a further optimization of the embodiment of the present application, in order to ensure the best performance of the fine-tuned large-scale language model in practical applications, a dual optimization strategy combining large model automatic evaluation and manual evaluation is adopted. First, the fine-tuned model is evaluated for performance using large model automatic evaluation, mainly by quantifying the performance of the model through indicators such as accuracy and rationality score of reasoning. Large model automatic evaluation can quickly identify the performance of the model under different conditions and provide guidance for preliminary optimization. Secondly, manual evaluation is performed to supplement the details that may be missed by automatic evaluation. Specifically, an evaluation team composed of experts in the field is formed to strictly review the output results of the model. Experts evaluate the performance of the model in actual application scenarios based on their professional knowledge, such as whether the reasoning process strictly implements the requirements of the regulations and whether the generated solution omits key links. Combined with the automatic evaluation results and expert rating results, the model is repeatedly iterated and optimized to gradually improve the accuracy and reliability of the model in practical applications.
[0191] Through the train operation risk identification method based on the large language model (LLM) proposed in the embodiment of the present application, a model TSLM specifically used for train risk monitoring is constructed through a hybrid enhancement strategy, that is, combining RAG and domain-specific fine-tuning. The performance of TSLM in various risk scenarios has high accuracy, reasonable reasoning process and effective risk disposal solutions. In untrained scenarios, TSLM can still maintain a high level of risk identification accuracy and rationality, demonstrating its strong generalization ability and reliability. The risk identification system framework provided by TSLM provides a new technical approach to provide more accurate data support for safety management for train scheduling safety assessment. Through continuous optimization and improvement, the TSLM model is expected to play an important role in more practical scenarios and further ensure the safety and efficiency of railway transportation.
[0192] Embodiment 5
[0193] Embodiment 5 of the present invention provides a non-volatile storage medium, wherein the storage medium stores at least one executable instruction, and the executable instruction can execute the virtual dialogue processing method in any of the above method embodiments.
[0194] The executable instructions may be used to enable the processor to execute various operations including those in the above-mentioned Embodiments 1 to 5.
[0195] Embodiment 6
[0196] Figure 5 A schematic diagram of the structure of a computing device according to Embodiment 4 of the present invention is shown. The specific embodiment of the present invention does not limit the specific implementation of the computing device.
[0197] like Figure 5 As shown, the computing device may include: a processor (processor) 702 , a communications interface (Communications Interface) 704 , a memory (memory) 706 , and a communication bus 708 .
[0198] in:
[0199] The processor 702 , the communication interface 704 , and the memory 706 communicate with each other via a communication bus 708 .
[0200] The communication interface 704 is used to communicate with other devices such as clients or other servers.
[0201] The processor 702 is used to execute the program 710, and specifically can execute the relevant steps in the above-mentioned virtual dialogue processing method embodiment.
[0202] Specifically, the program 710 may include program codes, which include computer operation instructions.
[0203] The processor 702 may be a central processing unit (CPU), or an application-specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of the present invention. The one or more processors included in the computing device may be processors of the same type, such as one or more CPUs; or processors of different types, such as one or more CPUs and one or more ASICs.
[0204] The memory 706 is used to store the program 710. The memory 706 may include a high-speed RAM memory, and may also include a non-volatile memory (non-volatile memory), such as at least one disk memory.
[0205] The program 710 may be specifically used to enable the processor 702 to execute various operations in the above-mentioned embodiments 1 to 5.
[0206] Through the solution provided by this embodiment, real-time monitoring and control data of the train are obtained, train dispatching operation log data are collected, and prompt instructions are generated based on the real-time monitoring and control data of the train and the train dispatching operation log data. Finally, the prompt instructions are sent to the large language model for train operation risk detection, and the output results of the large language model for train operation risk detection are obtained. The output results include risk identification results for train dispatching behavior. It can be seen that this embodiment provides a computing device that integrates safety risk identification technology for train dispatching operations with high intelligence and strong reliability, which can detect risks hidden in dispatching behavior in real time, and is of great value in preventing the occurrence of train operation safety problems.
[0207] The algorithm and display provided herein are not inherently related to any particular computer, virtual system or other device. Various general purpose systems can also be used together with the teachings based on this. According to the above description, it is obvious that the structure required for constructing such systems. In addition, the present invention is not directed to any specific programming language either. It should be understood that various programming languages can be utilized to realize the content of the present invention described herein, and the description of the above specific languages is for disclosing the best mode of the present invention.
[0208] In the description provided herein, a large number of specific details are described. However, it is understood that embodiments of the present invention can be practiced without these specific details. In some instances, well-known methods, structures and techniques are not shown in detail so as not to obscure the understanding of this description.
[0209] Similarly, it should be understood that in order to streamline the present disclosure and aid in understanding one or more of the various inventive aspects, in the above description of exemplary embodiments of the present invention, various features of the present invention are sometimes grouped together into a single embodiment, figure, or description thereof. However, this disclosed method should not be interpreted as reflecting the intention that the claimed invention requires more features than those expressly recited in each claim. Rather, as reflected in the claims, inventive aspects lie in less than all of the features of the individual embodiments previously disclosed. Therefore, the claims that follow the detailed description are hereby expressly incorporated into the detailed description, with each claim itself serving as a separate embodiment of the present invention.
[0210] Those skilled in the art will appreciate that the modules in the devices in the embodiments may be adaptively changed and arranged in one or more devices different from the embodiments. The modules or units or components in the embodiments may be combined into one module or unit or component, and in addition they may be divided into a plurality of submodules or subunits or subcomponents. Except that at least some of such features and / or processes or units are mutually exclusive, all features disclosed in this specification (including the accompanying claims, abstracts and drawings) and all processes or units of any method or device disclosed in this manner may be combined in any combination. Unless otherwise expressly stated, each feature disclosed in this specification (including the accompanying claims, abstracts and drawings) may be replaced by an alternative feature providing the same, equivalent or similar purpose.
[0211] In addition, those skilled in the art will appreciate that, although some embodiments described herein include certain features included in other embodiments but not other features, the combination of features of different embodiments is meant to be within the scope of the present invention and form different embodiments. For example, in the claims, any one of the claimed embodiments may be used in any combination.
[0212] It should be noted that the above embodiments illustrate the present invention rather than limit it, and that those skilled in the art may design alternative embodiments without departing from the scope of the appended claims. In the claims, any reference symbol between brackets should not be constructed as a limitation to the claims. The word "comprising" does not exclude the presence of elements or steps not listed in the claims. The word "one" or "an" preceding an element does not exclude the presence of a plurality of such elements. The present invention can be implemented by means of hardware including a number of different elements and by means of a suitably programmed computer. The use of the words first, second, and third, etc. does not indicate any order. These words can be interpreted as names.
Claims
1. A train operation risk identification method, comprising: Obtain real-time monitoring and control data of trains; Collect train dispatch operation log data; Generate prompt instructions based on the train real-time monitoring and control data and the train dispatching operation log data; The prompt instruction is sent to a large language model for train operation risk detection, and an output result of the large language model for train operation risk detection is obtained, wherein the output result includes a risk identification result for train scheduling behavior.
2. According to the method of claim 1, the train real-time monitoring and control data includes: Train control system data, and / or, train reception and dispatching plan data, and / or, real-time risk monitoring data.
3. According to the method of claim 1, the collecting of train dispatching operation log data comprises: Setting an agent module, wherein the agent module is configured to collect the train dispatching operation log data; The train dispatching operation log data includes: train dispatching log text data and train dispatching instruction audio data.
4. The method according to claim 1 or 3, wherein the collecting of train dispatching operation log data further comprises: Converting the train dispatching instruction audio data into second train dispatching instruction text data; The converting of the train dispatching instruction audio data into the second train dispatching instruction text data further includes: after converting the train dispatching instruction audio data into the first train dispatching instruction text data, formatting the first train dispatching instruction text data to obtain the formatted second train dispatching instruction text data.
5. The method according to claim 4, wherein the converting the train dispatching instruction audio data into the second train dispatching instruction text data further comprises: The train dispatching instruction audio data is converted into first train dispatching instruction text data using a speech recognition model, and the semantic extraction large language model is used to perform semantic extraction on the first train dispatching instruction text data to obtain the formatted second train dispatching instruction text data.
6. The method according to claim 1, generating a prompt instruction based on the train real-time monitoring and control data and the train dispatching operation log data comprises: The train real-time monitoring and control data and the train scheduling operation log data are configured as at least part of the input data in the prompt instruction, and are used as at least part of the data analysis basis for indicating the output of the execution result of the large language model for train operation risk detection.
7. The method according to claim 6, generating a prompt instruction based on the train real-time monitoring and control data and the train dispatching operation log data further comprises: The train real-time monitoring and control data and the train dispatching operation log data are configured to be used to retrieve and obtain text data matching the train real-time monitoring and control data and the train dispatching operation log data from a train dispatching related knowledge database, and the text data matching the train real-time monitoring and control data and the train dispatching operation log data are configured as at least part of the input data in the prompt instruction, and are used as at least part of the data analysis basis for indicating the output of the execution result of the large language model for train operation risk detection.
8. According to the method of claim 7, the train real-time monitoring and control data and the train dispatching operation log data are configured to be used to retrieve and obtain text data matching the train real-time monitoring and control data and the train dispatching operation log data from a train dispatching related knowledge database, including: The train scheduling related knowledge database includes an unstructured data vector database and / or a structured data database.
9. The method according to claim 7 or 8, before the train real-time monitoring and control data and the train dispatching operation log data are configured to be used to retrieve and obtain text data matching the train real-time monitoring and control data and the train dispatching operation log data from a train dispatching related knowledge database, further comprising: The train real-time monitoring and control data and the train dispatching operation log data are text-vectorized.
10. The method according to claim 1, wherein the prompt instruction is sent to a large language model for train operation risk detection, and an output result of the large language model for train operation risk detection is obtained, wherein the output result includes a risk identification result of a train scheduling behavior, including: The train operation risk detection large language model is a fine-tuned large language model.
11. The method according to claim 10, wherein the large language model for detecting train operation risk is a fine-tuned large language model, comprising: Using a few-sample thought chaining prompting method, a training dataset for the fine-tuning is generated.
12. The method according to claim 10 or 11, wherein the train operation risk detection large language model is a fine-tuned large language model, comprising: The fine tuning is performed using the LORA method.
13. The method according to any one of claims 1-3, 5-8, 10-11, after sending the prompt instruction to the large language model for train operation risk detection and obtaining the output result of the large language model for train operation risk detection, wherein the output result includes the risk identification result of the train scheduling behavior, comprises: The risk identification result includes the risk conclusion and the reasoning and analysis process for obtaining the risk conclusion.
14. A computing device comprising: A processor, a memory, a communication interface and a communication bus, wherein the processor, the memory and the communication interface communicate with each other via the communication bus; The memory is used to store at least one executable instruction, and the executable instruction enables the processor to perform the following operations: Obtain real-time monitoring and control data of trains; Collect train dispatch operation log data; Generate prompt instructions based on the train real-time monitoring and control data and the train dispatching operation log data; The prompt instruction is sent to a large language model for train operation risk detection, and an output result of the large language model for train operation risk detection is obtained, wherein the output result includes a risk identification result for train scheduling behavior.
15. A computer storage medium, wherein at least one executable instruction is stored in the storage medium, and the executable instruction enables a processor to execute an operation corresponding to a train operation risk identification method as described in any one of claims 1-13.
Citation Information
Patent Citations
Speech recognition method based on fine-tuning large language model in noisy environment
CN117423340A
Trackside auxiliary decision-making center equipment based on hybrid expert model
CN117864218A
Risk identification method, electronic device, storage medium and program product
CN117876090A
Rail transit equipment fault prediction method based on large language model
CN118133095A
Multi-modal data exception identification method and device, electronic equipment and storage medium
CN118379755A
Cited By
A traffic accident cause analysis method, device, equipment and medium
CN119760573A
A method, device, equipment and medium for analyzing the causes of traffic accidents
CN119760573B
