Urban rail dispatching command intelligent identification and emergency response method and system

Abstract

The invention provides an urban rail dispatching command intelligent identification and emergency response method and system, through the combination of a semantic retrieval technology and a large language model, the system can quickly and accurately retrieve documents related to faults from a local knowledge base, an emergency response instruction is generated in an intelligent mode, and specifically, the method and the system are convenient to use. Problems proposed by a user are matched with vector data in a local knowledge base, scheduling commands and emergency response suggestions based on a model are automatically generated, risks caused by human errors or delay are avoided, in addition, the efficiency and accuracy of fault emergency response are remarkably improved, the fault processing time is effectively shortened, and the user experience is improved. And the possibility of accidents is reduced.

Description

Technical Field

[0001] This invention belongs to the field of intelligent recognition and emergency response technology for urban rail transit dispatching commands, and specifically relates to a method and system for intelligent recognition and emergency response of urban rail transit dispatching commands. Background Technology

[0002] With the expansion of urban rail transit networks, system safety and emergency response capabilities have become pressing issues. Traditional emergency response methods rely on manual judgment, which carries risks of slow response times and operational errors, especially in extreme weather or special circumstances, potentially leading to lengthy processing times and significant losses. These traditional methods cannot meet the demands of modern rail transit for rapid response and efficient decision-making in the event of an emergency.

[0003] Existing technologies primarily rely on human intervention and a basic knowledge base, typically using expert-provided plans or rule bases for fault handling decisions. For example, knowledge graph-based emergency response methods can automatically generate plans, but they are still limited by the speed of rule updates and expert knowledge integration. This makes emergency response plans less flexible and real-time when facing complex and dynamically changing fault situations. Furthermore, traditional methods have high costs for knowledge base updates and management, and their emergency response effectiveness is easily affected by human factors, resulting in slow processing speeds. Summary of the Invention

[0004] Based on this, the present invention provides a method and system for intelligent identification and emergency response of urban rail dispatching commands, which aims to solve the problems of slow response speed, poor accuracy and operational errors caused by relying on manual judgment and traditional emergency handling methods in the prior art.

[0005] A first aspect of this invention provides a method for intelligent identification and emergency response of urban rail transit dispatching commands, the method comprising: Acquire train fault data, real-time operation data, and equipment status data, and establish a local knowledge base, which includes at least document data on train fault types, historical dispatching cases, and emergency handling procedures. The document data is converted into text data, and the text data is adaptively boundary-identified and split into several text paragraphs; The text paragraphs are converted into a first vector and stored in a vector database; When a train malfunctions, the text of the user's inquiry question is obtained and converted into a second vector. Based on the first vector and the second vector, semantic retrieval is performed to match and obtain the corresponding text paragraphs; The matched text paragraphs are combined with the preset dispatch instruction template to obtain the target consultation question text; Input the target consultation question text into the large language model, and output emergency response suggestions and dispatch commands related to the fault; The large language model employs a self-attention mechanism and includes at least a multi-head attention layer, a feedforward neural network, and a GLMBlock layer. The feedforward neural network uses the SwiGLU activation function.

[0006] Furthermore, the large language model includes an embedding layer, a first normalization layer, a multi-head attention layer, a residual connection and normalization hybrid layer, a feedforward neural network, a residual connection layer, a GLMBlock layer, a second normalization layer, and a fully connected layer, which are connected in sequence. The multi-head attention layer has 32 heads, and the output of the residual connection layer is calculated cyclically 28 times in the GLMBlock layer.

[0007] Furthermore, the feedforward neural network includes a first linear layer, a Swiglu activation function, and a second linear layer connected in sequence.

[0008] Furthermore, the step of inputting the target consultation question text into a large language model and outputting emergency response suggestions and dispatch commands related to the fault includes: The target consultation question text is converted into a machine-understandable form, and the continuous text is segmented into individual words or characters through word segmentation, and corresponding vector representations are generated. Randomly sample words or characters after text segmentation to generate multiple spans. Each span corresponds to a series of consecutive tokens in the input text. Mask the tokens to obtain the corrupted text portion. The order of multiple spans in the undamaged text portion is randomly shuffled, and the damaged text portion is concatenated with the undamaged text portion to obtain the final input text; The final input text is fed into a large language model, which outputs emergency response suggestions and dispatch commands related to the fault.

[0009] Furthermore, before the step of inputting the final input text into the large language model and outputting fault-related emergency response suggestions and dispatch commands, the following steps are included: The final input text is fed into the embedding layer. After transformation by high-dimensional word vectors and combination with positional information, an input matrix is ​​formed. Then, it enters the normalization processing layer and is normalized using RMSNorm.

[0010] Furthermore, in the step of converting the document data into text data, the document is loaded using the document_loaders class carried by LangChain.

[0011] Furthermore, the step of converting the document data into text data and performing adaptive boundary recognition on the text data to split it into several text paragraphs includes: The document data is converted into text data. The temporal boundary probability of each sentence in the text data is extracted by a step state machine based on LSTM. At the same time, the logical boundary probability of each sentence in the text data is extracted by a causal logic tree based on GNN. Based on the temporal boundary probability and the logical boundary probability, calculate the initial boundary score for each sentence of the text data, and mark the positions where the initial boundary score exceeds a first preset value as boundary candidate points; A contrastive learning model for the urban rail transit domain is adopted to convert each sentence into a contrastive semantic vector and to calculate the contrastive semantic distance between the sentences before and after the boundary candidate point. The contrastive semantic distance between the sentences before and after each of the aforementioned boundary candidate points is compared with the average difference value of the domain to determine enhancement candidate points. The enhancement score of each enhancement candidate point is calculated based on the initial boundary score, the contrastive semantic distance, and the average difference value of the domain. The text data is input into a pre-trained text classification model, which outputs the corresponding text type. Based on the text type, a threshold base value is called. DBSCAN clustering is performed on the contrastive semantic vectors of all sentences to calculate the overall topic density of the text. Based on the topic density, the threshold base value is adjusted to obtain the final threshold. Determine whether the enhanced score is greater than the final threshold; If so, determine the corresponding enhancement candidate point, calculate the domain entity overlap rate of the paragraphs before and after the corresponding enhancement candidate point, and determine whether the domain entity overlap rate is less than the second preset value. If so, the corresponding enhancement candidate point is determined as the final boundary, and the text data is split into several text segments based on the final boundary.

[0012] A second aspect of this invention provides an intelligent identification and emergency response system for urban rail transit dispatching commands, used to implement the intelligent identification and emergency response method for urban rail transit dispatching commands described in the first aspect, the system comprising: The acquisition module is used to acquire train fault data, real-time operation data, and equipment status data, and to establish a local knowledge base. The local knowledge base includes at least document data on train fault types, historical dispatching cases, and emergency handling procedures. The first conversion module is used to convert the document data into text data and perform adaptive boundary recognition on the text data to split it into several text paragraphs; The second conversion module is used to convert the text paragraph into a first vector and store it in a vector database; The third conversion module is used to obtain the text of the user's inquiry when the train malfunctions, and convert the inquiry text into a second vector. The retrieval module is used to perform semantic retrieval based on the first vector and the second vector, and to match the corresponding text paragraphs. The module combines the matched text paragraphs with a preset dispatch instruction template to obtain the target consultation question text. The input module is used to input the target consultation question text into the large language model and output emergency response suggestions and dispatch commands related to the fault. The large language model employs a self-attention mechanism and includes at least a multi-head attention layer, a feedforward neural network, and a GLMBlock layer. The feedforward neural network uses the SwiGLU activation function.

[0013] A third aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the intelligent identification and emergency response method for urban rail transit dispatching commands provided in the first aspect.

[0014] A fourth aspect of the present invention provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the intelligent identification and emergency response method for urban rail dispatching commands provided in the first aspect.

[0015] The present invention provides an intelligent identification and emergency response method and system for urban rail transit dispatching commands. By combining semantic retrieval technology with a large language model, the system can quickly and accurately retrieve fault-related documents from a local knowledge base and generate emergency response instructions in an intelligent manner. Specifically, it matches user-submitted questions with vector data in the local knowledge base to automatically generate model-based dispatching commands and emergency response suggestions, avoiding the risks caused by human error or delays. In addition, it significantly improves the efficiency and accuracy of fault emergency response, effectively shortens fault handling time, and reduces the possibility of accidents. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating the implementation of an intelligent identification and emergency response method for urban rail transit dispatching commands provided in Embodiment 1 of the present invention. Figure 2 This is a structural block diagram of an intelligent identification and emergency response system for urban rail dispatching commands provided in Embodiment 3 of the present invention; Figure 3This is a structural block diagram of an electronic device provided in Embodiment 4 of the present invention. Detailed Implementation

[0017] To facilitate understanding of the present invention, a more complete description will be given below with reference to the accompanying drawings. Several embodiments of the invention are illustrated in the drawings. However, the invention can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete.

[0018] It should be noted that when a component is said to be "fixed to" another component, it can be directly on the other component or there may be an intervening component. When a component is said to be "connected to" another component, it can be directly connected to the other component or there may be an intervening component. The terms "vertical," "horizontal," "left," "right," and similar expressions used in this document are for illustrative purposes only.

[0019] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0020] Example 1 According to an embodiment of the present invention, an embodiment of a method for intelligent identification and emergency response of urban rail transit dispatching commands is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0021] This first embodiment provides a method for intelligent identification and emergency response to urban rail transit dispatching commands, which can be used in electronic devices, such as computers. Please refer to... Figure 1 , Figure 1 The flowchart of the implementation of an intelligent identification and emergency response method for urban rail dispatching commands provided in Embodiment 1 of the present invention is shown, specifically including steps S01 to S07.

[0022] Step S01: Obtain train fault data, real-time operation data, and equipment status data, and establish a local knowledge base. The local knowledge base includes at least document data on train fault types, historical dispatching cases, and emergency handling procedures.

[0023] Specifically, the system collects train fault data and real-time operational data, including fault information, timetables, speed, position, acceleration, and other data. It also collects status data from related equipment (such as communication and signaling systems). This data is transmitted in real-time to storage devices via sensors and the data acquisition system, becoming the foundation for subsequent processing.

[0024] In addition, a local knowledge base is established, which includes at least document data on train fault types, historical dispatching cases, and emergency response procedures. This document data needs to be converted into text data to facilitate subsequent semantic retrieval. Understandably, the local knowledge base is established based on the acquired fault data, operational data, and equipment status data, and the document data on train fault types, historical dispatching cases, and emergency response procedures are compiled from this basic data.

[0025] Step S02: Convert the document data into text data, and perform adaptive boundary recognition on the text data to split it into several text paragraphs.

[0026] It should be noted that while traditional fixed-length text segmentation is convenient, it usually requires setting a relatively long fixed length, resulting in lengthy text content that is not conducive to subsequent text processing. Furthermore, fixed-length text segmentation may fragment semantics. To make text segmentation more reasonable and accurate, in this embodiment of the invention, the document data is first converted into text data. Then, using an LSTM-based step state machine, the temporal boundary probability of each sentence in the text data is extracted; a higher probability indicates a higher likelihood of a temporal boundary. Simultaneously, a GNN-based causal logic tree is used to extract the logical boundary probability of each sentence in the text data; a higher probability indicates a higher likelihood of a logical boundary. Based on the temporal boundary probability and the logical boundary probability, the initial boundary score for each sentence in the text data is calculated, and the positions corresponding to the initial boundary score exceeding a first preset value are marked as boundary candidate points. Specifically, the formula for calculating the initial boundary score is as follows: Initial boundary score = α × temporal boundary probability + (1-α) × logical boundary probability; α represents the weight. In this embodiment of the invention, α = 0.7 for process-type text and α = 0.3 for analysis-type text. A contrastive learning model for the urban rail transit domain is adopted to convert each sentence into a contrastive semantic vector and calculate the contrastive semantic distance between sentences before and after the boundary candidate point. In this embodiment of the invention, the contrastive learning model for the urban rail transit domain is a model based on SimCSE fine-tuning. The contrastive semantic distance can be expressed as: contrastive semantic distance = 1 - cosine similarity of contrastive semantic vectors. The larger the contrastive semantic distance, the more significant the topic difference. The contrastive semantic distance between the sentences before and after each of the aforementioned boundary candidate points is compared with the average difference value of the domain to determine the enhancement candidate points. In this embodiment of the invention, the average difference value of the domain is 0.5. If the contrastive semantic distance between the sentences before and after the boundary candidate point is greater than 0.5, then the boundary candidate point is determined as an enhancement candidate point. The enhancement score of the enhancement candidate point is calculated based on the initial boundary score, the contrastive semantic distance, and the average difference value of the domain. The expression is: Enhancement score = Initial boundary score × (1 + 0.5 × (Contrastive semantic distance - Average difference value of the domain)). The text data is input into a pre-trained text classification model, which outputs the corresponding text type. Based on the text type, a threshold base value is called. Specifically, the text classification model is a BERT classifier, which automatically identifies the text type, such as emergency procedures, fault cases, scheduling rules, etc. In addition, the threshold base value is obtained through historical data statistics. For example, the base value for the procedure class is 0.65, and the base value for the case class is 0.55. DBSCAN clustering is performed on the contrast semantic vectors of all sentences to calculate the overall topic density of the text. Based on the topic density, the threshold base value is adjusted to obtain the final threshold. It can be understood that if the topic density is high, that is, the sentence semantics are concentrated, such as in the standard operating procedure, the threshold is increased by 0.1 on the base value to reduce excessive segmentation. If the topic density is low, that is, the sentence semantics are scattered, such as in documents with multiple cases, the threshold is decreased by 0.1 to avoid missed detection boundaries. The final threshold can be expressed as: final threshold = threshold base value × (1 + topic density coefficient); Determine whether the enhanced score is greater than the final threshold; If so, the corresponding enhancement candidate point is determined, and the domain entity overlap rate of the paragraphs before and after the corresponding enhancement candidate point is calculated. It is then determined whether the domain entity overlap rate is less than the second preset value. The domain entity overlap rate can be understood as the proportion of common fault types and device names. If so, the corresponding enhancement candidate point is determined as the final boundary, and the text data is split into several text segments based on the final boundary.

[0027] Step S03: Convert the text paragraph into a first vector and store it in the vector database.

[0028] Specifically, the text embedding model text2vec3 is used to convert the split text into first vectors and store them in a vector database in Chroma format. Data in the knowledge base is preprocessed, converted into vectors, and stored in the vector database for subsequent semantic retrieval. The knowledge base can be built based on expert knowledge or through the continuous accumulation and updating of historical events and case data to ensure the timeliness and accuracy of emergency response plans.

[0029] Step S04: When a train malfunctions, obtain the text of the user's inquiry and convert the text of the inquiry into a second vector.

[0030] When a train malfunctions, users or dispatchers input relevant fault information or questions via natural language. The text embedding model text2vec3 is used to transform the user's input into a second vector.

[0031] Step S05: Based on the first vector and the second vector, perform semantic retrieval to match and obtain the corresponding text paragraphs.

[0032] In this embodiment of the invention, the first vector and the transformed second vector are semantically retrieved against text in a local knowledge base, and text segments with high relevance to the problem are selected through similarity calculation. This process involves natural language processing and vectorization techniques to ensure that the retrieved information is closely related to the current fault situation.

[0033] Step S06: Combine the matched text paragraphs with the preset dispatch instruction template to obtain the target consultation question text.

[0034] Specifically, based on the retrieved text and the predefined scheduling instruction template, combined with the user's input question, the final question is generated to be asked to the large language model.

[0035] Step S07: Input the target consultation question text into the large language model and output emergency response suggestions and dispatch commands related to the fault.

[0036] When the model receives a question, it first transforms it into a machine-understandable form. After word segmentation, the continuous text is divided into individual words or characters, and corresponding vector representations are generated. Next, the model randomly samples the input text, generating multiple spans. Each span corresponds to a series of consecutive tokens in the input text, which are then replaced with specific markers, generating a corrupted text portion (Part A). Then, the model randomly shuffles the order of multiple spans in Part B and concatenates Part A and Part B together. At this point, the text enters the embedding layer, where it undergoes high-dimensional word vector transformation and, combined with positional information, forms an input matrix. This matrix then enters the normalization layer, where RMSnorm is used for normalization.

[0037] In this embodiment of the invention, the large language model adopts a self-attention mechanism, including at least a multi-head attention layer, a feedforward neural network, and a GLMBlock layer, wherein the feedforward neural network uses the SwiGLU activation function.

[0038] Specifically, the input matrix is ​​processed using a self-attention mechanism to calculate the relevance of each word to other words, and a multi-head attention mechanism is used to compute the attention weight of each word in parallel to determine which other parts of the text it should pay attention to when generating the answer. Then, the model further extracts text features through a feed-forward network layer and uses the SwiGLU activation function to enhance the model's expressive power. The output of each layer is residually connected to the input to enhance the model's stability. After iterative computation through multiple GLMBlock layers, the model finally generates a set of probability vectors representing the probability of each possible output symbol. The symbol with the highest probability is then selected as the answer and output.

[0039] More specifically, the large language model includes, in sequence, an embedding layer, a first normalization layer, a multi-head attention layer, a residual connection and normalization hybrid layer, a feedforward neural network, a residual connection layer, a GLMBlock layer, a second normalization layer, and a fully connected layer. The multi-head attention layer has 32 heads, and the output of the residual connection layer is cyclically calculated 28 times in the GLMBlock layer. In addition, the feedforward neural network includes, in sequence, a first linear layer, a SwiGLU activation function, and a second linear layer.

[0040] This problem is addressed using a large language model, which generates emergency response suggestions and dispatch commands related to the fault. These can include train fault diagnosis, repair suggestions, and operational dispatch instructions, with the specific content depending on the fault type and emergency handling procedures.

[0041] In summary, the intelligent identification and emergency response method for urban rail dispatching commands in the above embodiments of the present invention combines semantic retrieval technology with a large language model, enabling the system to quickly and accurately retrieve fault-related documents from the local knowledge base and generate emergency response commands in an intelligent manner. Specifically, it matches user-submitted questions with vector data in the local knowledge base to automatically generate model-based dispatching commands and emergency response suggestions, avoiding the risks caused by human error or delays. In addition, it significantly improves the efficiency and accuracy of fault emergency response, effectively shortens fault handling time, and reduces the possibility of accidents.

[0042] Example 2 Embodiment 2 of the present invention provides a specific implementation method for intelligent identification and emergency response of urban rail dispatching commands, which is easy to understand in Embodiment 1 of the present invention. Specifically, when a train malfunctions, the user asks the model: "The problem is: the train derailed during operation. What should be the handling process in different time periods of 1-5 minutes, 5-20 minutes, and 20 minutes and above?" The model's answer is: Regarding the train derailment problem, the handling process at different time periods is as follows: 1-5 minutes: -The dispatcher should obtain the following information from the reporting person: 1) Train number and carriage number; 2) The exact location (specific kilometer marker) where the train derailed; 3) The train derailment situation, whether it obstructs adjacent lines and passenger capacity, etc.; 4) Are there any casualties?

[0043] - Take measures to impound trains to control train intervals and prevent trains from entering the affected sections. Suspend train service on the affected sections. For trains already in the affected sections, efforts should be made to guide them to the stations.

[0044] - Inform all stations along the line, as well as the drivers and crew dispatch offices of all trains on the line, of the actual situation.

[0045] - Assign the station manager on duty as the chief of accident handling and promptly notify the chief of accident handling on duty.

[0046] - Arrange for subsequent trains to remain at the platform or proceed to the nearest station as much as possible, and organize trains that stop between stations to run to the nearest station to avoid any trains being stranded between the two stations.

[0047] - Consult with the accident handling director on passenger evacuation routes and arrange for power outages on the contact rails in the relevant sections (according to the section evacuation plan).

[0048] - Stations that are unable to operate should be closed, and passengers on trains at the station should be cleared and put on standby.

[0049] - Notify the fault alarm center and arrange for various professional emergency response teams to arrive at the scene to assist in the rescue.

[0050] 5-20 minutes: - Take measures to impound trains to control train intervals and prevent trains from entering the affected sections. Suspend train service on the affected sections of the line.

[0051] - Make full use of the available conditions on the line and take measures such as clearing people and turning back midway as appropriate to maintain the operation of the section to the greatest extent possible.

[0052] -After the power supply is successfully restored, the train dispatcher organizes the trains to resume normal operation.

[0053] - If power supply fails, the system dispatcher determines the fault is caused by an external fault at the substation. If a single-sided or double-sided power supply method is used, the system dispatcher will notify the train dispatcher.

[0054] - Notify the accident handling director, the safety and quality department, and the public relations department.

[0055] - Notify TCC, TOCC, Public Transport Corps, and Emergency Response Office of the Transportation Commission.

[0056] 20 minutes and above: - The contact rail in the relevant section is de-energized. Trains stopped in the faulty section will be evacuated from the section. Station staff will be notified to assist.

[0057] - Make full use of the available conditions on the line, and take measures such as clearing passengers and turning back midway as needed to maintain the operation of the section to the greatest extent possible. Inform all stations and train drivers of the situation.

[0058] - Maintain close contact with the accident handling director, follow up on the development of the situation, grasp the situation on the ground, and adjust the traffic organization plan.

[0059] -After the emergency repairs are completed, confirm with the accident handling director whether the line has been cleared, whether the conditions for train operation are met, and the matters related to energizing the contact rail.

[0060] - Arrange for energizing the contact rail according to the actual situation.

[0061] -Resume normal operations and arrange the first train to conduct line inspection.

[0062] Example 3 Please see Figure 2 , Figure 2 This is a structural block diagram of an intelligent identification and emergency response system for urban rail transit dispatching commands provided in Embodiment 3 of the present invention. This intelligent identification and emergency response system 200 is used to implement the above embodiments and preferred embodiments; details already described will not be repeated. As used below, the term "module" can refer to a combination of software and / or hardware that performs a predetermined function. Although the apparatus described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0063] Specifically, the urban rail transit dispatch command intelligent identification and emergency response system 200 includes: an acquisition module 21, a first conversion module 22, a second conversion module 23, a third conversion module 24, a retrieval module 25, a combination module 26, and an input module 27, wherein: The acquisition module 21 is used to acquire train fault data, real-time operation data and equipment status data, and to establish a local knowledge base. The local knowledge base includes at least document data on train fault types, historical dispatch cases and emergency handling procedures. The first conversion module 22 is used to convert the document data into text data and perform adaptive boundary recognition on the text data to split it into several text paragraphs; The second conversion module 23 is used to convert the text paragraph into a first vector and store it in a vector database; The third conversion module 24 is used to acquire the text of the user's inquiry when the train malfunctions, and convert the text of the inquiry into a second vector; The retrieval module 25 is used to perform semantic retrieval based on the first vector and the second vector, and to match the corresponding text paragraphs. Module 26 is used to combine the matched text paragraphs with a preset dispatch instruction template to obtain the target consultation question text; Input module 27 is used to input the target consultation question text into the large language model and output emergency response suggestions and dispatch commands related to the fault. The large language model includes an embedding layer, a first normalization layer, a multi-head attention layer, a residual connection and normalization hybrid layer, a feedforward neural network, a residual connection layer, a GLMBlock layer, a second normalization layer, and a fully connected layer connected in sequence. The multi-head attention layer has 32 heads, and the output of the residual connection layer is calculated cyclically 28 times in the GLMBlock layer. The feedforward neural network includes a first linear layer, a SwiGLU activation function, and a second linear layer connected in sequence. The large language model employs a self-attention mechanism and includes at least a multi-head attention layer, a feedforward neural network, and a GLMBlock layer. The feedforward neural network uses the SwiGLU activation function.

[0064] Furthermore, in some other embodiments of the present invention, the input module 27 includes: The word segmentation processing unit is used to convert the target consultation question text into a machine-understandable form. After word segmentation processing, the continuous text is divided into individual words or characters, and corresponding vector representations are generated. The segmentation unit is used to randomly sample words or characters after text segmentation, generate multiple spans, each span corresponds to a series of consecutive tokens in the input text, and mask the tokens to obtain the damaged text portion; The concatenation unit is used to randomly shuffle the order of multiple spans in the undamaged text portion and concatenate the damaged text portion with the undamaged text portion to obtain the final input text; The input unit is used to input the final input text into the large language model and output emergency response suggestions and dispatch commands related to the fault.

[0065] Furthermore, in other embodiments of the present invention, the urban rail transit dispatching command intelligent identification and emergency response system 200 further includes: The normalization module is used to input the final input text into the embedding layer. After transforming the high-dimensional word vectors and combining them with positional information, an input matrix is ​​formed, which then enters the normalization processing layer and is normalized using RMSNorm.

[0066] Furthermore, in some other embodiments of the present invention, the first conversion module 22 includes: The extraction unit is used to convert the document data into text data, extract the temporal boundary probability of each sentence in the text data through an LSTM-based step state machine, and extract the logical boundary probability of each sentence in the text data through a GNN-based causal logic tree. The first calculation unit is used to calculate the initial boundary score of each sentence in the text data based on the temporal boundary probability and the logical boundary probability, and to mark the positions where the initial boundary score exceeds a first preset value as boundary candidate points. The second computing unit is used to adopt a contrastive learning model in the urban rail domain to convert each sentence into a contrastive semantic vector and calculate the contrastive semantic distance between the sentences before and after the boundary candidate point. The comparison unit is used to compare the contrast semantic distance between the sentences before and after each of the boundary candidate points with the average difference value of the domain, determine the enhancement candidate points, and calculate the enhancement score of the enhancement candidate points based on the initial boundary score, the contrast semantic distance and the average difference value of the domain. The calling unit is used to input the text data into a pre-trained text classification model, output the corresponding text type, and call the threshold base value according to the text type; The clustering unit is used to perform DBSCAN clustering on the contrastive semantic vectors of all sentences, calculate the overall topic density of the text, and adjust the threshold base value according to the topic density to obtain the final threshold. The first judgment unit is used to determine whether the enhanced score is greater than the final threshold. The second judgment unit is used to determine the corresponding enhancement candidate point when the enhancement score is greater than the final threshold, calculate the domain entity overlap rate of the paragraphs before and after the corresponding enhancement candidate point, and determine whether the domain entity overlap rate is less than a second preset value. The splitting unit is used to determine the corresponding enhancement candidate point as the final boundary when the overlap rate of the domain entities is less than a second preset value, and to split the text data into several text segments according to the final boundary.

[0067] Example 4 In another aspect, the present invention also proposes an electronic device, please refer to [link to relevant documentation]. Figure 3 The electronic device shown is an embodiment of the present invention, including a memory 20, a processor 10, and a computer program 30 stored in the memory and capable of running on the processor. When the processor 10 executes the computer program 30, it implements the intelligent identification and emergency response method for urban rail dispatching commands as described above.

[0068] In some embodiments, the processor 10 may be a central processing unit (CPU), controller, microcontroller, microprocessor or other data processing chip, used to run program code stored in memory 20 or process data, such as executing access restriction programs.

[0069] The memory 20 includes at least one type of readable storage medium, such as flash memory, hard disk, multimedia card, card-type memory (e.g., SD or DX memory), magnetic memory, magnetic disk, optical disk, etc. In some embodiments, the memory 20 can be an internal storage unit of an electronic device, such as the hard disk of the electronic device. In other embodiments, the memory 20 can also be an external storage device of the electronic device, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc. Furthermore, the memory 20 can include both internal and external storage units of the electronic device. The memory 20 can be used not only to store application software and various types of data of the electronic device, but also to temporarily store data that has been output or will be output.

[0070] It should be pointed out that, Figure 3 The structure shown does not constitute a limitation on the electronic device. In other embodiments, the electronic device may include fewer or more components than shown, or combine certain components, or have different component arrangements.

[0071] This invention also proposes a computer-readable storage medium storing a computer program that, when executed by a processor, implements the aforementioned method for intelligent identification and emergency response of urban rail transit dispatching commands.

[0072] Those skilled in the art will understand that the logic and / or steps represented in the flowcharts or otherwise described herein, for example, can be considered as a ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can mean any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0073] More specific examples of computer-readable media (a non-exhaustive list) include: electrical connections (electronic devices) having one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which the program can be printed, because the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.

[0074] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0075] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0076] The above embodiments merely illustrate several implementation methods of the present invention, and their descriptions are relatively specific and detailed, but they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these all fall within the protection scope of the present invention. Therefore, the protection scope of this patent should be determined by the appended claims.

Claims

1. A method for intelligent identification and emergency response of urban rail transit dispatching commands, characterized in that, The method includes: Acquire train fault data, real-time operation data, and equipment status data, and establish a local knowledge base, which includes at least document data on train fault types, historical dispatching cases, and emergency handling procedures. The document data is converted into text data, and the text data is adaptively boundary-identified and split into several text paragraphs; The text paragraphs are converted into a first vector and stored in a vector database; When a train malfunctions, the text of the user's inquiry question is obtained and converted into a second vector. Based on the first vector and the second vector, semantic retrieval is performed to match and obtain the corresponding text paragraphs; The matched text paragraphs are combined with the preset dispatch instruction template to obtain the target consultation question text; Input the target consultation question text into the large language model, and output emergency response suggestions and dispatch commands related to the fault; The large language model employs a self-attention mechanism and includes at least a multi-head attention layer, a feedforward neural network, and a GLMBlock layer. The feedforward neural network uses the SwiGLU activation function.

2. The intelligent identification and emergency response method for urban rail transit dispatching commands according to claim 1, characterized in that, The large language model includes, in sequence, an embedding layer, a first normalization layer, a multi-head attention layer, a residual connection and normalization hybrid layer, a feedforward neural network, a residual connection layer, a GLMBlock layer, a second normalization layer, and a fully connected layer. The multi-head attention layer has 32 heads, and the output of the residual connection layer is cyclically calculated 28 times in the GLMBlock layer.

3. The intelligent identification and emergency response method for urban rail transit dispatching commands according to claim 2, characterized in that, The feedforward neural network includes a first linear layer, a Swiglu activation function, and a second linear layer connected in sequence.

4. The intelligent identification and emergency response method for urban rail transit dispatching orders according to claim 3, characterized in that, The steps of inputting the target consultation question text into the large language model and outputting emergency response suggestions and dispatch commands related to the fault include: The target consultation question text is converted into a machine-understandable form, and the continuous text is segmented into individual words or characters through word segmentation, and corresponding vector representations are generated. Randomly sample words or characters after text segmentation to generate multiple spans. Each span corresponds to a series of consecutive tokens in the input text. Mask the tokens to obtain the corrupted text portion. The order of multiple spans in the undamaged text portion is randomly shuffled, and the damaged text portion is concatenated with the undamaged text portion to obtain the final input text; The final input text is fed into a large language model, which outputs emergency response suggestions and dispatch commands related to the fault.

5. The intelligent identification and emergency response method for urban rail transit dispatching orders according to claim 4, characterized in that, Before the step of inputting the final input text into the large language model and outputting fault-related emergency response suggestions and dispatch commands, the following steps are included: The final input text is fed into the embedding layer. After transformation by high-dimensional word vectors and combination with positional information, an input matrix is ​​formed. Then, it enters the normalization processing layer and is normalized using RMSNorm.

6. The intelligent identification and emergency response method for urban rail transit dispatching orders according to claim 5, characterized in that, In the step of converting the document data into text data, the document is loaded using the document_loaders class carried by LangChain.

7. The intelligent identification and emergency response method for urban rail transit dispatching orders according to claim 6, characterized in that, The steps of converting the document data into text data and performing adaptive boundary recognition on the text data to split it into several text paragraphs include: The document data is converted into text data. The temporal boundary probability of each sentence in the text data is extracted by a step state machine based on LSTM. At the same time, the logical boundary probability of each sentence in the text data is extracted by a causal logic tree based on GNN. Based on the temporal boundary probability and the logical boundary probability, calculate the initial boundary score for each sentence of the text data, and mark the positions where the initial boundary score exceeds a first preset value as boundary candidate points; A contrastive learning model for the urban rail transit domain is adopted to convert each sentence into a contrastive semantic vector and to calculate the contrastive semantic distance between the sentences before and after the boundary candidate point. The contrastive semantic distance between the sentences before and after each of the aforementioned boundary candidate points is compared with the average difference value of the domain to determine enhancement candidate points. The enhancement score of each enhancement candidate point is calculated based on the initial boundary score, the contrastive semantic distance, and the average difference value of the domain. The text data is input into a pre-trained text classification model, which outputs the corresponding text type. Based on the text type, a threshold base value is called. DBSCAN clustering is performed on the contrastive semantic vectors of all sentences to calculate the overall topic density of the text. Based on the topic density, the threshold base value is adjusted to obtain the final threshold. Determine whether the enhanced score is greater than the final threshold; If so, determine the corresponding enhancement candidate point, calculate the domain entity overlap rate of the paragraphs before and after the corresponding enhancement candidate point, and determine whether the domain entity overlap rate is less than the second preset value. If so, the corresponding enhancement candidate point is determined as the final boundary, and the text data is split into several text segments based on the final boundary.

8. A smart identification and emergency response system for urban rail transit dispatching commands, characterized in that, The system is used to implement the intelligent identification and emergency response method for urban rail transit dispatching orders as described in any one of claims 1-7, the system comprising: The acquisition module is used to acquire train fault data, real-time operation data, and equipment status data, and to establish a local knowledge base. The local knowledge base includes at least document data on train fault types, historical dispatching cases, and emergency handling procedures. The first conversion module is used to convert the document data into text data and perform adaptive boundary recognition on the text data to split it into several text paragraphs; The second conversion module is used to convert the text paragraph into a first vector and store it in a vector database; The third conversion module is used to obtain the text of the user's inquiry when the train malfunctions, and convert the inquiry text into a second vector. The retrieval module is used to perform semantic retrieval based on the first vector and the second vector, and to match the corresponding text paragraphs. The module combines the matched text paragraphs with a preset dispatch instruction template to obtain the target consultation question text. The input module is used to input the target consultation question text into the large language model and output emergency response suggestions and dispatch commands related to the fault. The large language model employs a self-attention mechanism and includes at least a multi-head attention layer, a feedforward neural network, and a GLMBlock layer. The feedforward neural network uses the SwiGLU activation function.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by the processor, the program implements the intelligent identification and emergency response method for urban rail transit dispatching commands as described in any one of claims 1-7.

10. An electronic device, characterized in that, It includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the intelligent identification and emergency response method for urban rail dispatching commands as described in any one of claims 1-7.

Images