Driving dispatcher training method, device and system based on voice model

Abstract

The invention relates to the technical field of urban rail transit technical training, and provides a driving dispatcher training method, device and system based on a voice model, and the method comprises the steps: triggering a corresponding fault after a student logs in, and loading a corresponding simulation emergency scene; analyzing the input information of the student through a voice model, and determining the type of the input information; if the type of the input information is an inquiry instruction, obtaining reply information; sending the reply information as feedback information to the trainee; if the type of the input information is an operation instruction, obtaining an operation result; the operation result is used as feedback information to be sent to the student; and if the training of the driving dispatcher is not finished, returning to execute the step of receiving the input information of the trainee. According to the scheme of the invention, an actual fault scene can be accurately simulated, personalized input of a student can be adapted, the ability of the student to flexibly deal with various complex scenes can be cultivated, and the flexibility and intelligence of simulated voice dialogue are improved.

Description

Technical Field

[0001] This invention relates to the field of urban rail transit technology training technology, and in particular to a method, device and system for training train dispatchers based on a voice model. Background Technology

[0002] Urban rail transit safety incidents are characterized by their suddenness, low probability of occurrence, but extremely high risk. For example, unexpected situations such as train malfunctions, power outages, and station fires can potentially paralyze the transportation system, cause massive train delays, and even lead to serious accidents like stampedes, severely endangering passengers' lives and property. Currently, with the continuous improvement of the stability of urban rail transit equipment and the increasingly sound regulations, the frequency of emergencies is gradually decreasing. However, this also means that urban rail transit staff have fewer opportunities to encounter emergencies in their daily work, resulting in a lack of experience in handling such situations. This situation presents new challenges for urban rail transit training. Therefore, the content and format of training for urban rail transit staff need to be more closely aligned with real-world scenarios, combining comprehensiveness, relevance, and flexibility. Systematic training should help staff accumulate experience in handling emergencies, thereby ensuring the continued safe and stable operation of urban rail transit.

[0003] Let's take the training methods for urban rail transit dispatchers as an example. Urban rail transit dispatchers occupy a central command position in the daily operation of rail transit, shouldering key responsibilities such as train operation organization, improving operational service quality, and ensuring transportation safety. Currently, the training methods for urban rail transit dispatchers mainly include two types: sand table simulation training and simulation software training.

[0004] In simulation-based training, trainees simulate different roles such as station staff, drivers, and superiors, recreating realistic fault scenarios through communication with dispatchers. Trainees convey fault information to dispatchers through dialogue, thus conducting fault handling training. This method requires multiple instructors working together, resulting in significant human and material resource investment and high training costs; moreover, the training effectiveness largely depends on the instructors' professional level. Furthermore, the training scenarios rely primarily on verbal descriptions from instructors, lacking visual presentation, making it difficult for dispatchers to gain a sense of immersion, leading to inconsistent training outcomes and a lack of consistent quality.

[0005] For simulation software training, technologies such as speech recognition, speech synthesis, and scene digitization effectively reduce the need for training support personnel, improving training effectiveness while lowering costs. Furthermore, simulation training software can utilize signal system simulation and 3D real-world modeling to construct more intuitive and realistic training scenarios. Trained dispatchers follow the scenario files generated by the scenario script, completing various operations sequentially. Upon inputting correct instructions at the prescribed steps, the system provides corresponding feedback based on the scenario file. However, current simulation software training heavily relies on scenario files, which often have only a single main flow and lack branching variations. This forces trainees to operate step-by-step, making it difficult to flexibly respond to different situations. Moreover, system feedback is based solely on preset rules in the scenario file, unable to dynamically adjust to individual trainee input, resulting in a significant gap between the software and the complexity and variability of actual fault scenarios.

[0006] This shows that in the existing urban rail transit dispatcher training system, the simulation scenarios differ significantly from actual fault scenarios, making it unable to adapt to the personalized input of trainees, and the flexibility of the simulated voice dialogue is insufficient. Summary of the Invention

[0007] This invention provides a method, device, and system for training train dispatchers based on a voice model, which addresses the shortcomings of existing technologies, such as the large gap between simulated scenarios and actual fault scenarios, the inability to adapt to the personalized input of trainees, and the insufficient flexibility of simulated voice dialogue.

[0008] This invention provides a method for training traffic dispatchers based on a speech model, comprising the following steps: After a student logs in, the corresponding fault is triggered, and the corresponding simulated emergency scenario is loaded. The system receives input information from students and analyzes it using a voice model to determine the type of input information, which includes inquiry instructions and operation instructions. If the type of the input information is an inquiry instruction, then the model corresponding to the inquiry instruction is invoked so that the model corresponding to the inquiry instruction performs query processing based on the input information to obtain response information. The response information includes processing suggestions, the basis for the suggestions, and the expected consequences after taking the processing suggestions. The response information is then sent to the student as feedback information. If the type of the input information is an operation instruction, then the model corresponding to the operation instruction is invoked so that the model corresponding to the operation instruction executes the operation corresponding to the input information and obtains the operation result. The operation result includes the operation instruction execution status, fault scenario information after successful operation execution, or the reason for operation failure and correction prompts. The operation result is sent to the student as feedback information. Determine whether the train dispatcher training has ended; if the train dispatcher training has not ended, return to the step of receiving the trainee's input information.

[0009] According to the present invention, a method for training traffic dispatchers based on a speech model includes the following: Invoking the model corresponding to the query instruction, so that the model corresponding to the query instruction performs query processing based on the input information to obtain response information, comprising: The input information is analyzed using a speech model to identify the student's intent and fault description information; The fault-related information is input into the fault diagnosis model so that the fault diagnosis model can diagnose the current fault and obtain the fault diagnosis result, which includes the fault type, the scope of impact, and the expected number of passengers stranded. Based on the fault diagnosis results and the trainee's intentions, the knowledge database is queried for handling procedures, safety regulations, and historical case provisions to obtain handling suggestions and the basis for those suggestions. The expected consequences of taking action recommendations are determined by using a traffic scheduling optimization algorithm.

[0010] According to the present invention, a traffic dispatcher training method based on a speech model is provided, wherein the step of calling the model corresponding to the operation instruction, so that the model corresponding to the operation instruction executes the operation corresponding to the input information, and obtaining the operation result includes: The input information is parsed using a speech model, and a high-order scheduling command is generated. The high-order scheduling command is decomposed into a sequence of operation instructions that can be directly issued to the interlocking system by applying a train scheduling optimization algorithm. Execute the sequence of operation instructions in a simulation environment; If the sequence of operation instructions is executed successfully, the execution status of the operation instructions is marked as successful, and the fault scenario information is updated in real time. The fault scenario information includes the train position, signal status, passenger flow distribution, and delay value. If the operation instruction sequence fails to execute, the operation instruction execution status is marked as execution failure, and the reason for failure and correction prompts are determined.

[0011] According to a speech model-based train dispatcher training method provided by the present invention, before receiving the trainee's input information, the method further includes: The working mode of the speech model-based train dispatcher training system is determined, and the working mode includes a training mode and an assessment mode. In the training mode, the trainee's input information includes inquiry instructions and operation instructions; In the assessment mode, the input information of the trainees includes operation instructions.

[0012] According to the present invention, a train dispatcher training method based on a speech model is provided, wherein the speech model includes: Collect operational data, dispatch command data, fault case data, and training feedback data; The collected data is preprocessed, and the preprocessing includes at least one of the following: data cleaning, data labeling, data transformation, and data partitioning; The initial speech model is trained and optimized based on the preprocessed data to obtain a well-trained speech model. The trained speech model is evaluated based on preset performance metrics.

[0013] According to the present invention, a traffic dispatcher training method based on a speech model is provided, wherein the feedback information comprises multiple candidate text responses and the probability of each candidate text response; the method further includes: The feedback information is displayed to the trainee according to the highest probability or a preset sampling strategy.

[0014] According to the present invention, a traffic dispatcher training method based on a speech model is provided, the method further includes: After the train dispatcher training is completed, the input and feedback information from the training process should be saved.

[0015] According to the present invention, a traffic dispatcher training method based on a speech model is provided, the method further includes: After the train dispatcher training is completed, the trainees' performance is quantitatively scored and qualitatively analyzed based on the input and feedback information during the training process and according to the preset evaluation criteria.

[0016] The present invention also provides a train dispatcher training device based on a speech model, comprising the following modules: The loading module is used to trigger the corresponding fault and load the corresponding simulated emergency scenario after the student logs in; The analysis module is used to receive input information from trainees and analyze the input information through a voice model to determine the type of input information, which includes inquiry instructions and operation instructions. The query module is used to call the model corresponding to the query instruction if the type of the input information is an inquiry instruction, so that the model corresponding to the inquiry instruction performs query processing based on the input information to obtain response information. The response information includes processing suggestions, the basis for the suggestions, and the expected consequences after taking the processing suggestions; and sends the response information as feedback information to the trainee. The operation module is used to, if the type of the input information is an operation instruction, call the model corresponding to the operation instruction so that the model corresponding to the operation instruction executes the operation corresponding to the input information and obtains the operation result. The operation result includes the operation instruction execution status, fault scenario information after successful operation execution or the reason for operation failure and correction prompts; and sends the operation result as feedback information to the trainee. The judgment module is used to determine whether the train dispatcher training has ended; if the train dispatcher training has not ended, it returns to the step of receiving the trainee's input information.

[0017] This invention also provides a traffic dispatcher training system based on a speech model, comprising: The system consists of a second data acquisition layer, a second data preprocessing layer, a data storage layer, a model and algorithm layer, and a training and application layer. The second data acquisition layer is used to collect train operation data, equipment status data, passenger flow data, and student operation data; The second data preprocessing layer is used to preprocess the train operation data, the equipment status data, the passenger flow data, and the student operation data. The preprocessing includes at least one of the following: data cleaning, data labeling, data conversion, and data segmentation. The data storage layer includes an operation database, a scheduling business database, a student database, and a knowledge database. The data storage layer is used to store pre-processed train operation data, equipment status data, passenger flow data, and student operation data according to their types. The model and algorithm layer includes a speech model, a fault diagnosis model, a traffic dispatching optimization algorithm, and a student evaluation model. The model and algorithm layer is used to analyze the student's input information, diagnose faults, generate operation instruction sequences, return feedback information, and evaluate the student's performance. The training application layer includes a training course module, an interactive interface module, an intelligent interaction module, and an assessment module. The training application layer is used to provide a visual interactive interface, display fault scenarios and feedback information in real time during the training process. The training application layer is also used to save the input and feedback information during the training process after the train dispatcher training ends, and to perform quantitative scoring and qualitative analysis of the trainees' performance based on the input and feedback information during the training process and according to preset assessment standards.

[0018] The present invention provides a speech model-based method, apparatus, and system for training train dispatchers. This system analyzes trainee input information using a speech model, matches the corresponding model to the type of input information, processes it accordingly, and obtains feedback. It is understood that a speech model trained on a large scale can understand and process user commands of different forms and expressions. Dispatchers can freely communicate with the train dispatcher training system using natural language, inquiring about fault handling methods, querying dispatch order details, discussing strategies for handling special situations, or issuing operational instructions based on the current fault situation. The system can flexibly analyze the input information, execute corresponding operations, and provide accurate responses, without being limited by fixed procedures or preset questions. This is closer to the diverse and uncertain scenarios faced by dispatchers in actual work, helping to cultivate dispatchers' ability to flexibly cope with various complex situations. Therefore, the solution of this application can accurately simulate actual fault scenarios, adapt to personalized trainee input, help cultivate trainees' ability to flexibly cope with various complex scenarios, and improve the flexibility and intelligence of simulated voice dialogue. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0020] Figure 1 This is a flowchart illustrating the speech model-based train dispatcher training method provided by the present invention.

[0021] Figure 2 This is a data structure diagram of the speech model provided by the present invention.

[0022] Figure 3 This is a data structure diagram of the training system provided by the present invention.

[0023] Figure 4 This is a schematic diagram of the structure of the traffic dispatcher training device based on a voice model provided by the present invention. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0025] The following is combined with Figures 1-3 This invention describes a speech model-based method for training train dispatchers.

[0026] Figure 1 This is a flowchart illustrating the speech model-based train dispatcher training method provided by the present invention, as shown below. Figure 1 As shown, the method includes the following: Step 101: After the student logs in, trigger the corresponding fault and load the corresponding simulated emergency scenario; Step 102: Receive the input information from the student and analyze the input information through the voice model to determine the type of input information. The types of input information include inquiry instructions and operation instructions. Step 103: If the type of input information is an inquiry instruction, then call the model corresponding to the inquiry instruction so that the model corresponding to the inquiry instruction can perform query processing based on the input information and obtain response information. The response information includes processing suggestions, the basis for the suggestions, and the expected consequences after taking the processing suggestions. The response information is sent to the trainee as feedback information. Step 104: If the type of input information is an operation instruction, then call the model corresponding to the operation instruction so that the model corresponding to the operation instruction can execute the operation corresponding to the input information and obtain the operation result. The operation result includes the operation instruction execution status, the fault scenario information after successful operation execution or the reason for operation failure and correction prompts; send the operation result as feedback information to the trainee. Step 105: Determine whether the train dispatcher training has ended; Step 106: If the train dispatcher training is not yet completed, return to step 102.

[0027] In practical applications, the entity executing this speech model-based train dispatcher training method can be a speech model-based train dispatcher training device. There are various ways to implement a speech model-based train dispatcher training device, such as a computer server or a cloud platform.

[0028] Specifically, step 101 includes: after the student logs in, triggering the corresponding fault and loading the corresponding simulated emergency scenario.

[0029] In practical applications, trainees can log in to the speech model-based train dispatcher training system via terminal devices. For example, the speech model-based train dispatcher training system can be either training software or a training platform.

[0030] In this application, no specific limitation is made on the terminal device; for example, the terminal device can be a computer, a server, etc. In one example, the trainee enters their account and verification information on the login interface. After the voice model-based traffic dispatcher training system verifies the trainee's identity, the trainee logs in successfully.

[0031] The simulated emergency scenario refers to a virtual, real-world-like environment provided by the voice model-based train dispatcher training system to simulate specific situations when a fault occurs. The scenario information for the simulated emergency scenario is detailed through text descriptions, relevant charts, simple animations, and voice announcements, specifying key information such as the time, location, and scope of impact of the fault.

[0032] This application leverages the rich knowledge base and generative capabilities of speech models to create a variety of complex emergency scenarios, such as single-fault scenarios like train derailment, signal system malfunctions, and sudden fires in stations, as well as composite scenarios involving multiple simultaneous faults. These scenarios are presented to trainees through text descriptions and voice announcements, enabling them to face various potential emergency situations in a virtual environment.

[0033] In practical applications, after a trainee logs in, the voice model-based train dispatcher training device automatically triggers the corresponding fault and loads the corresponding simulated emergency scenario. In one example, the voice model-based train dispatcher training system stores the trainee's training records. After logging in, based on the trainee's training records and the preset training plan, the system can determine the content the trainee currently needs to learn, automatically triggering the corresponding fault and loading the corresponding simulated emergency scenario. In another example, after logging in, the trainee can select the scenario topic they need to learn about. Based on the user-selected scenario topic, the system automatically triggers the corresponding fault and loads the corresponding simulated emergency scenario.

[0034] Preferably, the speech model-based train dispatcher training system of this application also supports user selection of work mode. In one possible implementation, before step 102, the above-mentioned speech model-based train dispatcher training method further includes: The working mode of the speech model-based train dispatcher training system is determined, which includes a training mode and an assessment mode. In the training model, trainees' input information includes inquiry instructions and operation instructions; In the assessment mode, the trainees' input information includes operation instructions.

[0035] In practical applications, the train dispatcher training system in this application supports trainees in selecting training and assessment modes, meaning that trainees can be trained or assessed through the train dispatcher training system.

[0036] Specifically, in the training mode, trainees can use the input interface of the dispatcher training system to ask questions, issue instructions, or seek advice from the chatbot via text or voice input. For example, a dispatcher can ask, "Should we prioritize evacuating passengers?"

[0037] Specifically, in the assessment mode, the inquiry function is turned off. Trainees analyze the fault situation based on the simulated emergency scenario and issue operation instructions to the chatbot through the input interface of the train dispatcher training system by means of text input or voice input.

[0038] Further, step 102 includes: receiving input information from the student, and analyzing the input information from the student through a voice model to determine the type of input information, which includes inquiry instructions and operation instructions.

[0039] The speech model can be a model built on natural language processing (NLP) technology and trained on a large scale. For example, an initial speech model is built based on NLP technology, and professional knowledge such as urban rail transit dispatch instructions and standard fault handling procedures are used as data input to conduct large-scale training to obtain the speech model.

[0040] Optionally, in one possible implementation, the method further includes: Collect operational data, dispatch command data, fault case data, and training feedback data; The collected data is preprocessed, and the preprocessing includes at least one of the following: data cleaning, data labeling, data transformation, and data partitioning; The initial speech model is trained and optimized based on the preprocessed data to obtain a well-trained speech model. The trained speech model is evaluated based on preset performance metrics.

[0041] Specifically, Figure 2 This is a data structure diagram of the speech model provided by the present invention, such as... Figure 2 As shown, the speech model includes: a first data acquisition layer 21, a first data preprocessing layer 22, a model training layer 23, a model evaluation layer 24, and a model application layer 25.

[0042] The first data acquisition layer 21 is used to collect operational data, scheduling command data, fault case data, and training feedback data.

[0043] For example, the process begins with professional text embedding. Specifically, professional texts such as urban rail transit dispatching orders and standard fault handling procedures are converted into low-dimensional vectors using word embedding or character embedding techniques. For instance, professional terms such as "train reversal" and "station passenger evacuation" are mapped to specific vectors. Further, scene feature encoding is performed. Specifically, simulated emergency scenario information, such as the location of train malfunctions and fire locations, is encoded. These scene features are then converted into vector form and combined with the professional text vectors as input to the speech model.

[0044] For the first data acquisition layer 21, a domain adaptation layer is added on top of the common Transformer encoder to capture specific semantic and syntactic patterns in the urban rail transit scheduling domain. For example, it learns the sequence of different instructions in scheduling commands and the logical relationships in fault handling procedures.

[0045] Combination Figure 2 The first data preprocessing layer 22 is used to preprocess the data collected by the first data acquisition layer. The preprocessing includes at least one of the following: data cleaning, data labeling, data transformation and data partitioning.

[0046] Combination Figure 2 The model training layer 23 is used to build the initial speech model and to train and optimize the model based on the preprocessed data to obtain the trained speech model.

[0047] Optionally, the model training layer 23 can be a professional knowledge enhancement layer based on a multilayer perceptron (MLP). This layer utilizes a professional knowledge graph of the urban rail transit domain to further process and enhance the features extracted by the encoder, enabling the model to better understand scheduling-related professional concepts and relationships. For example, it clarifies the causal relationship between different fault types and corresponding handling measures.

[0048] Combination Figure 2 The model evaluation layer 24 is used to evaluate the speech model according to preset performance indicators. The model application layer 25 is used for training system inheritance, real-time feedback, and data updates.

[0049] Specifically, in model application layer 25, for situations requiring simulated scheduling operations, the decoder can also generate operation instruction sequences, such as control instructions for analog signal systems and train operation adjustment schemes, and verify the feasibility and effectiveness of the operations by interacting with the virtual scheduling environment.

[0050] In this embodiment, when the speech model performs word mapping for specialized text, the Transformer uses positional encoding to enable the model to perceive the positional information of the words. The main formula of the algorithm is as follows: In the formula, PE here refers to the position. pos Hewei i Combined, use cos , sin The function encodes positions in odd and even dimensions by dividing by 10000, thus encoding the positional information so that different positions have different encoded values. The total dimension of the language model determines the feature dimensions of all sub-layers in the entire Transformer network.

[0051] The attention mechanisms in the Transformer architecture specifically include ScaledDot-Product Attention and Multi-Head Attention. The formula for ScaledDot-Product Attention is as follows: in, The dimensions of the query and key vectors.

[0052] For the multi-head attention mechanism, the formula is as follows: in, To output the linear transformation matrix.

[0053] Based on the formulas of the scaled dot product attention mechanism and the multi-head attention mechanism, this application projects the input into multiple low-dimensional subspaces according to the input data, then calculates the attention in parallel, and finally concatenates the results and performs a linear transformation.

[0054] Furthermore, based on the scaled dot product attention, a mask matrix is ​​added. M To ensure that each position can only focus on the positions preceding it when calculating attention, the expression is as follows: in, In calculating the query ( Q The similarity score matrix between ) and key (K), To prevent the gradient from vanishing due to excessively large dot product results during scaling operations, the mask matrix is... M Adding it to the scoring matrix suppresses the score for future positions. softmax The scores are converted into a probability distribution. Due to the presence of the mask matrix, the scores for future positions are mapped to probabilities close to 0. Finally, attention weights are applied to the value matrix. V The output is obtained from the above.

[0055] Understandably, after training, the voice model can flexibly parse various commands issued by users, output corresponding responses based on the input content, and perform corresponding operations, providing intelligent interactive support for the training of urban rail transit dispatchers.

[0056] In practical applications, the speech model analyzes the student's input information to determine the type of input information. Based on the type of input information, the corresponding model is matched for appropriate processing. For query commands, step 103 is used for processing to obtain feedback information. For operation commands, step 104 is used for processing to obtain feedback information.

[0057] Specifically, step 103 includes: if the type of input information is an inquiry instruction, then calling the model corresponding to the inquiry instruction so that the model corresponding to the inquiry instruction performs query processing based on the input information to obtain response information, the response information including processing suggestions, the basis for the suggestions, and the expected consequences after taking the processing suggestions; and sending the response information as feedback information to the trainee.

[0058] Optionally, in one possible implementation, step 103 above calls the model corresponding to the query instruction so that the model corresponding to the query instruction performs query processing based on the input information to obtain response information, including: The input information is analyzed using a speech model to identify the student's intent and fault description information; Input the fault-related information into the fault diagnosis model so that the fault diagnosis model can diagnose the current fault and obtain the fault diagnosis results, including the fault type, the scope of impact, and the expected number of passengers stranded. Based on the fault diagnosis results and the trainee's intentions, the knowledge database is consulted for handling procedures, safety regulations, and historical case provisions to obtain handling suggestions and the basis for those suggestions; The expected consequences of taking action recommendations are determined by using a traffic scheduling optimization algorithm.

[0059] For example, the trainee's input is a query command, specifically, "Train A is on fire at platform 3, what should I do?" The speech model recognizes that the trainee's intention is to seek advice on how to handle the situation and extracts the fault description information, including "Train A", "platform 3", and "fire".

[0060] Furthermore, the fault diagnosis model outputs diagnostic results based on the input fault information. The diagnostic results include a diagnosis type of station fire, an affected area of ​​platform 3 and adjacent sections, and an estimated passenger backlog of 1200 people.

[0061] Furthermore, based on the fault diagnosis results and the trainees' intentions, the system queries the knowledge database for handling procedures, safety regulations, and historical case provisions to obtain handling suggestions and their basis. The handling suggestions include: immediately initiating the passenger evacuation procedure, based on Article 5.2.1 of the "Fire Emergency Response Code." Further, the system uses a traffic scheduling optimization algorithm to determine the expected consequences of implementing the handling suggestions, with an estimated passenger evacuation time of 6 minutes and an estimated delay of 3 minutes.

[0062] Furthermore, the handling recommendations, their basis, and the expected consequences of implementing them are integrated into a response message: "It is recommended to immediately initiate the passenger evacuation procedure for Platform 3 (in accordance with Article 5.2.1 of the 'Fire Emergency Response Code'), which is expected to be completed in 6 minutes, resulting in a 3-minute delay." This response message is then sent to the trainees as feedback.

[0063] In this implementation, the system uses a voice model to analyze trainees' intentions, providing personalized feedback and enhancing their learning experience and training effectiveness. Through a fault diagnosis model and knowledge database, the system provides accurate fault diagnosis results and handling suggestions, helping trainees quickly understand the fault situation and take the correct action. The traffic dispatch optimization algorithm, based on real-time data and historical cases, provides scientific decision support, ensuring trainees learn the optimal handling methods during training. The system can generate detailed handling suggestions and justifications based on trainee input, helping trainees better understand the decision-making process and improving training quality. Therefore, this implementation significantly improves the efficiency and quality of traffic dispatcher training while reducing training costs and enhancing the trainees' learning experience.

[0064] Specifically, step 104 includes: if the type of input information is an operation instruction, then call the model corresponding to the operation instruction so that the model corresponding to the operation instruction can execute the operation corresponding to the input information and obtain the operation result. The operation result includes the operation instruction execution status, fault scenario information after successful operation execution, or the reason for operation failure and correction prompt; and send the operation result as feedback information to the trainee.

[0065] Optionally, in one possible implementation, step 104 above involves calling the model corresponding to the operation instruction to cause the model corresponding to the operation instruction to perform the operation corresponding to the input information and obtain the operation result, including: The input information is parsed using a speech model, and a high-order scheduling command is generated. The high-order scheduling command is decomposed into a sequence of operation instructions that can be directly issued to the interlocking system by applying a train scheduling optimization algorithm. Execute the sequence of operation instructions in a simulation environment; If the sequence of operation instructions is executed successfully, the execution status of the operation instructions will be marked as successful, and the fault scenario information will be updated in real time. The fault scenario information includes the train position, signal status, passenger flow distribution and delay value. If the operation instruction sequence fails to execute, the operation instruction execution status will be marked as execution failure, and the reason for the failure and correction prompts will be determined.

[0066] For example, the student's input is an operation command, specifically "Close the uplink signal S1101 on platform 3". The speech model recognizes that the student's intention is to execute an operation command and generates a higher-order dispatch command: "Close the uplink signal S1101 on platform 3".

[0067] Furthermore, by applying the train dispatching optimization algorithm, the high-order dispatching command is decomposed into the following sequence of operation instructions: Close signal S1101, sending the instruction to close signal S1101 to the interlocking system; Set track section occupancy, setting the up track section of platform 3 to "occupancy" to prevent trains from entering; Notify subsequent trains, sending deceleration or stopping instructions to subsequent trains to ensure safety.

[0068] Furthermore, after receiving the sequence of operation instructions, the simulation environment executes each instruction in sequence: shutting down signal S1101, setting track section occupancy, and notifying subsequent trains to slow down or stop.

[0069] Furthermore, if the sequence of operation instructions is executed successfully, the execution status is marked as successful, and the fault scenario information is updated in real time. The fault scenario information after successful operation execution includes: train position, signal status, passenger flow distribution, and delay value. Specifically, the train position is: the following train slows down or stops outside the station; the signal status is: signal S1101 displays red; the passenger flow distribution is: the passenger congestion situation on platform 3 is updated; and the delay value is: the delay time of the following train is updated to increase by 3 minutes.

[0070] Furthermore, if the operation instruction fails to execute, the operation instruction execution status is marked as execution failure, and the reason for the failure and correction prompts are determined.

[0071] Furthermore, if the operation command fails to execute, the execution status, reason for failure, and modification prompts are combined into the operation result; if the operation command executes successfully, the execution status and fault scenario information after successful execution are combined into the operation result. The operation result is then sent as feedback information to the trainee.

[0072] In this implementation, through a voice model and a train dispatching optimization algorithm, the system can automatically break down high-level dispatching commands into directly executable sequences of operational instructions, reducing the complexity and error rate of manual operations. The simulation environment can execute operational instructions in real time and update fault scenario information, ensuring trainees can immediately see the results, enhancing the realism and interactivity of the training. The system can identify the reasons for operational instruction execution failures and provide correction prompts, helping trainees learn how to handle problems in actual operations and improving their emergency response capabilities. Through automated operational instruction processing and real-time feedback, trainees gain more practical opportunities during training, improving their operational proficiency and decision-making abilities. Therefore, this implementation significantly improves the efficiency and quality of train dispatcher training while reducing training costs and enhancing the trainees' learning experience.

[0073] Optionally, in one possible implementation, the feedback information includes multiple candidate text responses and the probability of each candidate text response; the method further includes: Feedback information is displayed to trainees based on the highest probability or a preset sampling strategy.

[0074] Alternatively, in one possible implementation, feedback information can be presented to trainees in a visual manner, such as by displaying a route map after the train operation adjustment and a simulation animation of station passenger flow evacuation through a graphical interface, to help trainees understand the training content and the effectiveness of their decisions more intuitively.

[0075] Step 105 includes: determining whether the train dispatcher training has ended. Step 106 includes: if the train dispatcher training has not ended, then return to step 102.

[0076] In practical applications, trainees can interact with the chatbot multiple times, asking questions, issuing instructions, or seeking advice. Therefore, after sending feedback to the trainee, the system determines whether the dispatcher training is complete. If the training is not complete, it continues to wait for input from the trainee.

[0077] Specifically, this application does not specify the exact method for determining whether train dispatcher training has ended. In one example, a maximum training duration for train dispatchers is pre-set. The system checks if the current training duration has reached the maximum; if so, the training is considered complete; otherwise, it is considered incomplete. In another example, the system checks if the trainee is logged in; if so, the training is considered incomplete; otherwise, it is considered complete.

[0078] Optionally, in one possible implementation, the above method further includes: After the train dispatcher training is completed, the input and feedback information from the training process should be saved.

[0079] In this implementation, after the train dispatcher training concludes, the input and feedback information from the training process is saved, providing a complete training record. This allows trainees and training managers to easily trace every step and decision made during the training. Trainees can review their operations and system feedback, analyze the rationality of their decisions, and thus better summarize lessons learned, thereby improving the effectiveness of the training.

[0080] Optionally, in one possible implementation, the above method further includes: After the train dispatcher training is completed, the trainees' performance is quantitatively scored and qualitatively analyzed based on the input and feedback information during the training process and according to the preset evaluation criteria.

[0081] For example, the preset evaluation criteria include, but are not limited to: decision-making accuracy, response timeliness, operational standardization, and the rationality of emergency strategies.

[0082] In this implementation, after the training, based on the input and feedback information collected during the training process, the trainees' performance is quantitatively scored and qualitatively analyzed according to preset evaluation criteria. This provides trainees with a comprehensive assessment and personalized feedback, helping them better understand their performance and develop improvement plans. These evaluation results not only contribute to the personal growth of trainees but also provide data support for training managers to optimize training content and methods, thereby improving the overall effectiveness and quality of training.

[0083] In addition, this application also provides a traffic dispatcher training system based on a speech model, for example, Figure 3 This is a data structure diagram of the training system provided by the present invention, such as... Figure 3 As shown, the traffic dispatcher training system based on the voice model includes: a second data acquisition layer 31, a second data preprocessing layer 32, a data storage layer 33, a model and algorithm layer 34, and a training application layer 35.

[0084] The second data acquisition layer 31 is used to collect train operation data, equipment status data, passenger flow data, and student operation data.

[0085] The second data preprocessing layer 32 is used to preprocess train operation data, equipment status data, passenger flow data and student operation data. The preprocessing includes at least one of the following: data cleaning, data labeling, data conversion and data segmentation.

[0086] The data storage layer 33 includes an operation database, a scheduling business database, a student database, and a knowledge database. The data storage layer 33 is used to store preprocessed train operation data, equipment status data, passenger flow data, and student operation data according to their types.

[0087] Specifically, the operations database stores data related to the daily operation of urban rail transit, such as train operation data (train timetables, train operation status, delay information, etc.), line topology (line layout, station information, section division, etc.), and equipment status (status information of signaling systems, track equipment, communication equipment, etc.). For example, it provides the initial status of trains and equipment during simulated emergency scenarios. During fault diagnosis and handling, it provides real-time train operation and equipment status data.

[0088] Specifically, the dispatching business database stores data related to dispatching operations, such as dispatching commands (various commands issued by dispatchers and their execution results), fault records (historical fault information, including fault type, time of occurrence, location, and scope of impact), and interlocking operation records (operation records executed by the interlocking system, such as signal switchover). For example, in a fault diagnosis model, historical fault data is provided for model training. During training, dispatcher operations and system feedback are recorded for subsequent evaluation and analysis.

[0089] Specifically, the trainee database stores data related to trainees. This includes trainee identification information, training records, and assessment scores. For example, trainee identity is verified upon login. After training, trainee performance is recorded and evaluation reports are generated. This provides training managers with trainee training and assessment data for management and decision-making purposes.

[0090] Specifically, the knowledge database stores professional knowledge and standards related to scheduling operations. Examples include handling procedures, safety regulations, and historical case provisions. For instance, it provides standards and guidelines when generating handling recommendations. During training, it provides trainees with detailed background knowledge and handling recommendations. In assessment models, it serves as a reference standard for evaluating trainee performance.

[0091] In practical applications, the operations database and scheduling database provide real-time data to support fault diagnosis and scenario simulation. The trainee database records trainee training and assessment data, supporting personalized learning and evaluation. The knowledge database provides professional knowledge and standards, supporting the generation and evaluation of processing suggestions. Through the collaborative work of these four databases, the system can provide a comprehensive, efficient, and personalized training and assessment environment to help trainees improve their ability to cope with emergencies.

[0092] Combination Figure 3 The model and algorithm layer 34 includes a speech model, a fault diagnosis model, a traffic scheduling optimization algorithm, and a student evaluation model. The model and algorithm layer 34 is used to analyze the student's input information, diagnose faults, generate operation instruction sequences, return feedback information, and evaluate the student's performance.

[0093] Specifically, the speech model is used for trainees' natural language input and generates natural language feedback. In training mode, trainees can ask for instructions via voice or text, and the speech model will analyze the input and generate detailed processing suggestions. In assessment mode, trainees can input operation instructions via voice or text, and the speech model will analyze the input and generate operation result feedback.

[0094] Specifically, the fault diagnosis model is used to diagnose the current fault and output detailed fault information. When the student inputs an inquiry command, the fault diagnosis model provides detailed fault diagnosis results to help the student understand the current situation. When the student inputs an operation command, the fault diagnosis model provides fault background information, supporting the speech model in generating reasonable operation suggestions.

[0095] Specifically, the train dispatching optimization algorithm decomposes high-order dispatching commands into directly executable sequences of operation instructions and calculates the operation results. When a student inputs an operation instruction, the algorithm generates a specific sequence of operation instructions and executes it in the simulation environment. When a student inputs a query instruction, the algorithm calculates the expected effect of the processing suggestion and supports the speech model in generating detailed processing suggestions.

[0096] Specifically, the trainee assessment model is used to quantitatively score and qualitatively analyze trainees' performance. After training, the model generates an assessment report based on trainees' input and feedback. This report helps trainees understand their performance and develop improvement plans. Furthermore, it serves as an important basis for trainee evaluation and certification.

[0097] Understandably, the speech model parses the trainee's input commands and generates natural language feedback. The fault diagnosis model provides detailed fault diagnosis results and supports the speech model in generating processing suggestions. The traffic dispatch optimization algorithm breaks down high-order commands into executable operation command sequences and calculates the operation results. The trainee evaluation model assesses the trainee's performance after training and generates a detailed evaluation report. Through the collaborative work of these four modules, the model and algorithm layer 34 can provide a comprehensive, efficient, and personalized training and assessment environment to help trainees improve their ability to cope with emergencies.

[0098] Combination Figure 3 The training application layer 35 includes a training course module, an interactive interface module, an intelligent interaction module, and an assessment module. The training application layer 35 is used to provide a visual interactive interface, display fault scenarios and feedback information in real time during the training process. The training application layer 35 is also used to save the input and feedback information during the training process after the train dispatcher training ends, and to perform quantitative scoring and qualitative analysis of the trainees' performance based on the input and feedback information during the training process and according to the preset assessment criteria.

[0099] Specifically, the training course module is responsible for managing and loading training course content. In practical applications, after logging in, trainees select a specific training topic through the training course module. The system loads the corresponding simulated emergency scenarios based on the trainee's selection and begins the training process. In this application, the training course module supports multiple training topics, such as train malfunction handling, fire emergencies, and signal system failures. Trainees can independently select the topic they need to train on, or use a preset training plan, and the corresponding training course will be automatically loaded.

[0100] Specifically, the interactive interface module provides a platform for trainees to interact with the system. In practical applications, trainees can ask questions or issue operation commands to the system through text or voice input. The system provides trainees with processing suggestions, operation results, and other information through text, voice feedback, charts, and animations. The interactive interface module also visualizes information such as fault scenarios, train operation status, signal status, and passenger flow distribution through a graphical interface. For example, during training, trainees input commands through the interactive interface module, and the system provides feedback through this module. The system updates fault scenario information in real time through the interactive interface module to help trainees better understand the current situation.

[0101] Specifically, the intelligent interaction module is used to enable intelligent interaction with trainees. In practical applications, it parses the trainee's natural language input, understands the trainee's intent, and generates corresponding responses. The interaction method is adjusted according to the system's operating mode (training mode or assessment mode). In training mode, trainees can ask questions to the system through the intelligent interaction module and obtain processing suggestions and evidence. In assessment mode, trainees can only issue operation commands through the intelligent interaction module, and the system provides feedback on the operation results.

[0102] Specifically, the assessment module is used to quantitatively score and qualitatively analyze trainees' performance. In practical application, the module scores trainees' performance numerically based on preset assessment criteria. These criteria may include decision-making accuracy, response timeliness, operational standardization, and the appropriateness of emergency strategies. The module provides a detailed descriptive analysis of trainees' performance, highlighting their strengths and weaknesses during the training process. Furthermore, the module generates detailed assessment reports, including trainees' training records, quantitative scores, and qualitative analysis results.

[0103] Understandably, the training course module loads training courses and simulated emergency scenarios to provide training content for trainees. The interactive interface module provides a user interface for interacting with the system, supporting trainees to input commands and receive system feedback. The intelligent interaction module enables intelligent interaction with trainees, adjusting the interaction method according to the work mode and supporting voice and natural language processing. The assessment module evaluates trainees' performance after training and generates detailed assessment reports. Through the collaborative work of these four modules, the training application layer 35 can provide a comprehensive, efficient, and personalized training and assessment environment to help trainees improve their ability to cope with emergencies.

[0104] To illustrate the speech model-based train dispatcher training method provided in this application, a specific example will be used below.

[0105] First, trainees log into the training software and enter the training interface. The software will automatically present the current simulated emergency scenario according to the preset training plan. The scenario information may include text descriptions, relevant charts, or simple animations, detailing key information such as the time, location, and scope of impact of the fault.

[0106] Furthermore, trainees can use the software's input interface to ask questions, issue commands, or seek advice from the chatbot through text or voice input. For example, a dispatcher could ask, "Should we prioritize evacuating passengers?"

[0107] Furthermore, the chatbot uses a pre-trained speech model to analyze the dispatcher's input intent, and then combines this with professional dispatching knowledge and emergency response procedures to provide real-time feedback on the software interface. It not only offers suggestions on whether to evacuate passengers, but also explains the relevant rationale and potential consequences.

[0108] Furthermore, based on the feedback from the chatbot, trainees can further reflect and make decisions, continuing to interact with the robot, such as asking, "If passengers are evacuated, what specific steps should be taken?" The robot will continuously guide dispatchers to consider the details and impacts of various decisions, helping them to form a complete emergency response plan.

[0109] Furthermore, during the interaction, the software generates corresponding dynamic changes in the scenario based on the dispatcher's decisions. For example, if the dispatcher makes a mistake or causes a delay, the software may display information such as the train delay time being extended and passengers in the station becoming more anxious, prompting the dispatcher to adjust their strategy in a timely manner.

[0110] Furthermore, for every operation and decision made by the dispatcher, the chatbot in the software not only judges its correctness but also provides in-depth explanations of the underlying principles and related knowledge. For example, when the dispatcher issues a dispatch instruction, the chatbot will explain in detail the connection between the instruction and safety regulations and operational rules, as well as the differences in the applicability of the instruction in different similar scenarios.

[0111] Furthermore, the software automatically records the interaction data between the dispatcher and the system, including the content of questions asked, decision-making time, and the sequence of instructions issued. After the training, based on preset evaluation criteria, the dispatcher's performance is quantitatively scored and qualitatively analyzed from multiple dimensions, including decision-making accuracy, response timeliness, operational standardization, and the rationality of emergency strategies.

[0112] The speech model-based train dispatcher training method provided by this invention analyzes the trainee's input information using a speech model, matches the corresponding model to the type of input information, and processes it accordingly to obtain feedback information. It is understood that a speech model obtained through large-scale training can understand and process user commands of different forms and expressions. Dispatchers can freely communicate with the train dispatcher training system using natural language, inquiring about fault handling methods, querying dispatch command details, discussing coping strategies for special situations, or issuing operation commands based on the current fault situation. The system can flexibly analyze the input information, execute corresponding operations, and provide accurate responses, without being limited by fixed procedures or preset questions. This is closer to the diverse and uncertain scenarios faced by dispatchers in actual work, helping to cultivate dispatchers' ability to flexibly cope with various complex situations. Therefore, the solution of this application can accurately simulate actual fault scenarios, adapt to the personalized input of trainees, help cultivate trainees' ability to flexibly cope with various complex scenarios, and improve the flexibility and intelligence of simulated voice dialogue.

[0113] The speech model-based train dispatcher training device provided by the present invention will be described below. The speech model-based train dispatcher training device described below can be referred to in correspondence with the speech model-based train dispatcher training method described above.

[0114] Figure 4 This is a schematic diagram of the structure of the speech model-based train dispatcher training device provided by the present invention, as shown below. Figure 4 As shown, the speech model-based train dispatcher training device includes: Loading module 41 is used to trigger the corresponding fault and load the corresponding simulated emergency scenario after the student logs in; Analysis module 42 is used to receive input information from students and analyze the input information through a voice model to determine the type of input information, which includes inquiry instructions and operation instructions. The query module 43 is used to call the model corresponding to the query instruction if the type of input information is a query instruction, so that the model corresponding to the query instruction can perform query processing based on the input information to obtain response information. The response information includes processing suggestions, the basis for the suggestions, and the expected consequences after taking the processing suggestions; and the response information is sent to the trainee as feedback information. Operation module 44 is used to call the model corresponding to the operation instruction if the type of input information is an operation instruction, so that the model corresponding to the operation instruction can execute the operation corresponding to the input information and obtain the operation result. The operation result includes the operation instruction execution status, the fault scenario information after successful operation execution or the reason for operation failure and correction prompts; and send the operation result as feedback information to the trainee. The judgment module 45 is used to determine whether the train dispatcher training has ended; if the train dispatcher training has not ended, it returns to the step of receiving the trainee's input information.

[0115] Preferably, in one possible implementation, the query module 43 is used to invoke the model corresponding to the query instruction, so that the model corresponding to the query instruction performs query processing based on the input information and obtains the response information. Specifically, it is used for: The input information is analyzed using a speech model to identify the student's intent and fault description information; Input the fault-related information into the fault diagnosis model so that the fault diagnosis model can diagnose the current fault and obtain the fault diagnosis results, including the fault type, the scope of impact, and the expected number of passengers stranded. Based on the fault diagnosis results and the trainee's intentions, the knowledge database is consulted for handling procedures, safety regulations, and historical case provisions to obtain handling suggestions and the basis for those suggestions; The expected consequences of taking action recommendations are determined by using a traffic scheduling optimization algorithm.

[0116] Preferably, in one possible implementation, the operation module 44 is used to call the model corresponding to the operation instruction so that the model corresponding to the operation instruction executes the operation corresponding to the input information and obtains the operation result, specifically for: The input information is parsed using a speech model, and a high-order scheduling command is generated. The high-order scheduling command is decomposed into a sequence of operation instructions that can be directly issued to the interlocking system by applying a train scheduling optimization algorithm. Execute the sequence of operation instructions in a simulation environment; If the sequence of operation instructions is executed successfully, the execution status of the operation instructions will be marked as successful, and the fault scenario information will be updated in real time. The fault scenario information includes the train position, signal status, passenger flow distribution and delay value. If the operation instruction sequence fails to execute, the operation instruction execution status will be marked as execution failure, and the reason for the failure and correction prompts will be determined.

[0117] Preferably, in one possible implementation, the above-mentioned speech model-based dispatcher training device further includes: a pattern determination module; the pattern determination module is used for: The working mode of the speech model-based train dispatcher training system is determined, which includes a training mode and an assessment mode. In the training model, trainees' input information includes inquiry instructions and operation instructions; In the assessment mode, the trainees' input information includes operation instructions.

[0118] Preferably, in one possible implementation, the above-described apparatus further includes: a training module; the training module is used for: Collect operational data, dispatch command data, fault case data, and training feedback data; The collected data is preprocessed, and the preprocessing includes at least one of the following: data cleaning, data labeling, data transformation, and data partitioning; The initial speech model is trained and optimized based on the preprocessed data to obtain a well-trained speech model. The trained speech model is evaluated based on preset performance metrics.

[0119] Optionally, in one possible implementation, the feedback information includes multiple candidate text responses and the probability of each candidate text response; the device further includes: The display module is used to show feedback information to trainees based on the highest probability or a preset sampling strategy.

[0120] Optionally, in one possible implementation, the above-described apparatus further includes: The storage module is used to save the input and feedback information during the training process after the train dispatcher training is completed.

[0121] Optionally, in one possible implementation, the above-described apparatus further includes: The assessment module is used to quantitatively score and qualitatively analyze trainees' performance based on input and feedback information during the training process, according to preset assessment criteria, after the train dispatcher training is completed.

[0122] The speech model-based train dispatcher training device provided by this invention analyzes the trainee's input information through a speech model, matches the corresponding model according to the type of input information, processes it accordingly, and obtains feedback information. It is understood that a speech model obtained through large-scale training can understand and process user commands of different forms and expressions. Dispatchers can freely communicate with the train dispatcher training system using natural language, inquiring about fault handling methods, querying dispatch command details, discussing coping strategies for special situations, or issuing operation commands based on the current fault situation. The system can flexibly analyze the input information, execute corresponding operations, and provide accurate responses, without being limited by fixed procedures or preset questions. This is closer to the diverse and uncertain scenarios faced by dispatchers in actual work, helping to cultivate dispatchers' ability to flexibly cope with various complex situations. Therefore, the solution of this application can accurately simulate actual fault scenarios, adapt to the personalized input of trainees, help cultivate trainees' ability to flexibly cope with various complex scenarios, and improve the flexibility and intelligence of simulated voice dialogue.

[0123] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0124] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0125] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A voice model based method for training a train dispatcher, the method comprising: The method comprises the following steps: After the student logs in, the corresponding fault is triggered, and the corresponding simulated emergency scene is loaded; Receiving input information of the student, and analyzing the input information of the student through a voice model to determine the type of the input information, the type of the input information including inquiry instructions and operation instructions; If the type of the input information is inquiry instructions, the model corresponding to the inquiry instructions is called to make the model corresponding to the inquiry instructions query processing based on the input information to obtain reply information, the reply information including processing suggestions, suggestion basis and expected consequences after taking the processing suggestions; The reply information is sent to the student as feedback information; If the type of the input information is operation instructions, the model corresponding to the operation instructions is called to make the model corresponding to the operation instructions execute the operation corresponding to the input information to obtain operation results, the operation results including operation instruction execution situation, fault scene information after operation execution success or operation execution failure reason and correction prompt; the operation results are sent to the student as feedback information; Determine whether the train dispatching officer training is completed; if the train dispatching officer training is not completed, return to execute the step of receiving the input information of the student.

2. The voice model based en route dispatcher training method of claim 1, wherein, The calling of the model corresponding to the inquiry instructions to make the model corresponding to the inquiry instructions query processing based on the input information to obtain reply information comprises: Analyzing the content of the input information through a voice model to identify the student's intention and fault description information; Inputting fault related information into a fault diagnosis model to make the fault diagnosis model diagnose the current fault to obtain fault diagnosis results, the fault diagnosis results including fault type, influence range and expected passenger retention amount; According to the fault diagnosis results and the student's intention, querying the disposal specification, safety regulations and historical case provisions in the knowledge database to obtain processing suggestions and suggestion basis; Determine the expected consequences after taking the processing suggestions through a train dispatching optimization algorithm.

3. The voice model based en route dispatcher training method of claim 1, wherein, The calling of the model corresponding to the operation instructions to make the model corresponding to the operation instructions execute the operation corresponding to the input information to obtain operation results comprises: Analyzing the content of the input information through a voice model and generating high-level dispatching commands; Applying a train dispatching optimization algorithm to decompose the high-level dispatching commands into operation instruction sequences that can be directly issued to interlocking; Executing the operation instruction sequences in a simulation environment; If the operation instruction sequences are executed successfully, the operation instruction execution situation is marked as successful execution, and the fault scene information is updated in real time, the fault scene information including train position, signal machine state, passenger flow distribution and late number value; If the operation instruction sequences fail to execute, the operation instruction execution situation is marked as failed execution, and the failure reason and correction prompt are determined.

4. The voice model based en route dispatcher training method of claim 1, wherein, Before receiving the input information of the student, the method further comprises: Determine the working mode of the train dispatching officer training system based on the voice model, the working mode including a training mode and an examination mode; In the training mode, the input information of the student includes inquiry instructions and operation instructions; In the examination mode, the input information of the student includes operation instructions.

5. The voice model based en route dispatcher training method of claim 1, wherein, The method further comprises: Collect operation data, dispatching command data, fault case data and training feedback data; Preprocess the collected data, the preprocessing including at least one of data cleaning, data labeling, data conversion and data division; Train and optimize the initial voice model based on the preprocessed data to obtain a trained voice model; Evaluate the trained voice model according to a preset performance index.

6. The voice model based en route dispatcher training method of claim 1, wherein, The feedback information includes multiple candidate text replies and probabilities of each candidate text reply; The method further includes: Display the feedback information to the trainee according to the highest probability or a preset sampling strategy.

7. The voice model based en route dispatcher training method according to any one of claims 1 to 6, characterized in that, The method further includes: After the trainee training ends, save the input information and feedback information during the training.

8. The voice model based en route dispatcher training method according to any one of claims 1-6, wherein, The method further includes: After the trainee training ends, quantitatively score and qualitatively analyze the performance of the trainee based on the input information and feedback information during the training according to a preset evaluation standard.

9. A voice model based trainer for a train dispatcher, characterized by The method further includes: Load the corresponding fault after the trainee logs in, and load the corresponding simulated emergency scene; Receive the input information of the trainee, and analyze the input information of the trainee through the voice model to determine the type of the input information, the type of the input information including inquiry instructions and operation instructions; If the type of the input information is inquiry instructions, call the model corresponding to the inquiry instructions to make the model corresponding to the inquiry instructions query and process based on the input information to obtain reply information, the reply information including processing suggestions, suggestion basis and expected consequences after taking the processing suggestions; Send the reply information to the trainee as feedback information; If the type of the input information is operation instructions, call the model corresponding to the operation instructions to make the model corresponding to the operation instructions execute the operation corresponding to the input information to obtain operation results, the operation results including operation instruction execution situation, fault scene information after operation execution success or operation execution failure reason and correction prompt; send the operation results to the trainee as feedback information; Determine whether the trainee training ends; if the trainee training does not end, return to the step of receiving the input information of the trainee.

10. A voice model based train dispatcher training system, comprising: The method further includes: The second data collection layer, the second data preprocessing layer, the data storage layer, the model and algorithm layer and the training application layer; The second data collection layer is configured to collect train operation data, equipment state data, passenger flow data and trainee operation data; The second data preprocessing layer is configured to preprocess the train operation data, the equipment state data, the passenger flow data and the trainee operation data, the preprocessing including at least one of data cleaning, data labeling, data conversion and data division; The data storage layer includes an operation database, a dispatching business database, a trainee database and a knowledge database, and is configured to store the preprocessed train operation data, equipment state data, passenger flow data and trainee operation data according to types. The model and algorithm layer includes a voice model, a fault diagnosis model, a driving dispatch optimization algorithm, and a trainee evaluation model, and is configured to analyze input information of the trainee, diagnose faults, generate an operation instruction sequence, return feedback information, and evaluate performance of the trainee; The training application layer includes a training course module, an interactive interface module, an intelligent interaction module, and an examination and evaluation module, and is configured to provide a visual interactive interface, display a fault scene and feedback information in real time during the training process; and save input information and feedback information during the training process when the training of the driving dispatcher is completed, and quantitatively score and qualitatively analyze performance of the trainee according to the input information and the feedback information during the training process and according to preset evaluation standards.

Images