Large-model-based drilling situation influence analysis method and system and storage medium

Abstract

The invention provides a drill situation influence analysis method and system based on a large model and a storage medium, and the method comprises the steps: determining a target station and a target end sealing region according to a drill instruction, obtaining the region image of each station region in the target station, and determining the people flow distribution information according to the region image; determining a people flow trend change according to the ticket selling information; determining drilling external information according to the station position of the target station and the current time; and inputting the target end sealing region, the people flow distribution information, the people flow trend change and the drilling external information into a pre-trained large model for situation influence analysis to obtain a drilling situation influence analysis result. According to the embodiment of the invention, the method employs a large model mode, carries out the drilling situation impact analysis of the target end sealing region according to the people flow distribution information, the people flow trend change, the drilling weather information, the drilling holiday information and the drilling event information, and improves the comprehensiveness and accuracy of the drilling situation impact analysis.

Description

Technical Field

[0001] The present invention relates to the field of intelligent transportation technology, and in particular to a large-model-based exercise situation impact analysis method, system and storage medium. Background Art

[0002] Station closure drills are drills that simulate the evacuation of personnel, safety closures, emergency rescue and other processes when a station encounters an emergency (such as fire, natural disasters, etc.). The drills are designed to improve the emergency response capabilities of station staff and passengers in emergencies, ensure personnel safety, and reduce casualties and property losses. During station closure drills, in order to ensure passenger travel efficiency, the issue of drill situation impact analysis of the drill plan has received increasing attention.

[0003] In the existing exercise situation impact analysis process, the exercise situation impact analysis of the exercise plan is generally performed based on manual experience, resulting in low accuracy of the exercise situation impact analysis. Summary of the invention

[0004] The purpose of the embodiments of the present invention is to provide a large-model-based exercise situation impact analysis method, system and storage medium to solve the problem of low accuracy of exercise situation impact analysis in the prior art.

[0005] The embodiment of the present invention is implemented as follows: a method for analyzing the impact of a drill situation based on a large model, the method comprising:

[0006] Obtaining a drill instruction, and determining a target station and a target blocking area according to the drill instruction;

[0007] Acquire a regional image of each station area in the target station, and determine crowd flow distribution information according to the regional image;

[0008] Acquire ticket sales information within a preset time period at the target station, and determine changes in passenger flow trends based on the ticket sales information;

[0009] Determining drill external information according to the station location and current time of the target station, the drill external information including drill weather information, drill holiday information and drill event information;

[0010] The target closed area, the crowd distribution information, the crowd trend change and the drill external information are input into the pre-trained large model to perform situation impact analysis to obtain the drill situation impact analysis result.

[0011] Preferably, before inputting the target closed area, the crowd flow distribution information, the crowd flow trend change and the drill external information into the pre-trained large model for situation impact analysis, the method further includes:

[0012] Acquire drill samples, and input the drill samples into the large model for crowd flow analysis to obtain drill internal sample information;

[0013] According to the drill weather samples, drill holiday samples and drill event samples in the drill samples, an impact prediction is performed on the closed sample area to obtain the external predicted impact of the drill;

[0014] Performing an impact prediction on the capped sample area according to the drill internal sample information to obtain the drill internal prediction impact, and determining the drill prediction situation according to the drill internal prediction impact and the drill external prediction impact;

[0015] The model loss is determined according to the internal prediction impact of the exercise, the external prediction impact of the exercise and the prediction situation of the exercise, and the parameters of the large model are updated according to the model loss to obtain the pre-trained large model.

[0016] Preferably, the drill sample is input into the large model for crowd flow analysis to obtain drill internal sample information, including:

[0017] Performing face recognition on the regional image of the sample area in the drill sample according to the large model to obtain the number of regional faces, and determining the crowd flow distribution sample value of the sample area according to the number of regional faces;

[0018] Acquire the ticketing sample information in the drill sample, and determine the target platform according to the user destination in the ticketing sample information;

[0019] Determine the number of changes in the flow of people in the sample area according to the target station, and determine the regional flow of people trend in the sample area according to the number of changes in the flow of people;

[0020] The drill internal sample information includes the regional passenger flow trend and the regional passenger flow distribution.

[0021] Preferably, determining the model loss according to the predicted internal impact of the exercise, the predicted external impact of the exercise and the predicted situation of the exercise includes:

[0022] Obtaining the real internal impact of the drill, the real external impact of the drill, and the real situation of the drill of the drill sample, and calculating the first loss according to the real internal impact of the drill and the predicted internal impact of the drill;

[0023] Calculating a second loss according to the actual external impact of the drill and the predicted external impact of the drill, and calculating a third loss according to the actual situation of the drill and the predicted situation of the drill;

[0024] Weight values ​​are calculated for the first loss, the second loss, and the third loss to obtain the model loss.

[0025] Preferably, the target closed area, the crowd flow distribution information, the crowd flow trend change and the drill external information are input into the pre-trained large model for situation impact analysis, and after obtaining the drill situation impact analysis result, the method further includes:

[0026] Input the target closed area and the drill external information into the pre-trained large model to perform drill area matching, obtain a first candidate drill plan, and obtain the historical drill situation of the first candidate drill plan;

[0027] Setting the first candidate exercise plan corresponding to the minimum historical exercise situation as the first optimized exercise plan;

[0028] Input the current time and the drill external information into the pre-trained large model to match the drill time, obtain a second candidate drill plan, and obtain the historical drill situation of the second candidate drill plan;

[0029] The second candidate exercise plan corresponding to the minimum historical exercise situation is set as a second optimized exercise plan, and plan optimization information is generated according to the first optimized exercise plan and the second optimized exercise plan.

[0030] Preferably, after obtaining the drill sample, the method further includes:

[0031] Noise sampling is performed according to the large model to obtain sampling noise, and sample generation is performed according to the sampling noise and the training sample to obtain a generated sample;

[0032] Performing sample judgment on the generated sample to obtain a judgment result, and determining the sample generation loss according to the judgment result;

[0033] updating parameters of the large model according to the sample generation loss until the large model converges, and inputting the training sample into the converged large model for data enhancement to obtain enhanced data;

[0034] The training sample is replaced by the enhanced data.

[0035] Preferably, the target closed area, the crowd flow distribution information, the crowd flow trend change and the drill external information are input into the pre-trained large model for situation impact analysis, and after obtaining the drill situation impact analysis result, the method further includes:

[0036] Determine a drill similarity scheme according to the target closed area, the crowd flow distribution information, the crowd flow trend change and the drill external information, and obtain drill problems and emergency plans for the drill similarity scheme;

[0037] A drill prompt is generated based on the drill question and the emergency plan.

[0038] Another object of an embodiment of the present invention is to provide a large model-based exercise situation impact analysis system, the system comprising:

[0039] An area determination module is used to obtain a drill instruction and determine a target station and a target end-blocking area according to the drill instruction;

[0040] A crowd flow analysis module, used to obtain a regional image of each station area in the target station, and determine crowd flow distribution information according to the regional image;

[0041] Acquire ticket sales information within a preset time period at the target station, and determine changes in passenger flow trends based on the ticket sales information;

[0042] An external analysis module, used to determine drill external information according to the station location and current time of the target station, wherein the drill external information includes drill weather information, drill holiday information and drill event information;

[0043] The situation analysis module is used to input the target closed area, the crowd distribution information, the crowd trend change and the drill external information into the pre-trained large model to perform situation impact analysis and obtain the drill situation impact analysis results.

[0044] Preferably, the exercise situation impact analysis system further includes:

[0045] The model training module is used to obtain drill samples and input the drill samples into the large model for crowd flow analysis to obtain drill internal sample information;

[0046] According to the drill weather samples, drill holiday samples and drill event samples in the drill samples, an impact prediction is performed on the closed sample area to obtain the external predicted impact of the drill;

[0047] Performing an impact prediction on the capped sample area according to the drill internal sample information to obtain the drill internal prediction impact, and determining the drill prediction situation according to the drill internal prediction impact and the drill external prediction impact;

[0048] The model loss is determined according to the internal prediction impact of the exercise, the external prediction impact of the exercise and the prediction situation of the exercise, and the parameters of the large model are updated according to the model loss to obtain the pre-trained large model.

[0049] The embodiment of the present invention adopts a large model approach to perform a drill situation impact analysis on the target closed area based on crowd distribution information, crowd trend changes, drill weather information, drill holiday information, and drill event information, thereby improving the comprehensiveness and accuracy of the drill situation impact analysis and enabling the drill situation impact analysis results to more accurately reflect the actual situation. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 is a flow chart of a method for analyzing the impact of a drill situation based on a large model provided in a first embodiment of the present invention;

[0051] Figure 2 It is a structural diagram of a large model-based exercise situation impact analysis system provided by a second embodiment of the present invention;

[0052] Figure 3 It is a schematic diagram of the framework of the exercise situation impact analysis system based on a large model provided by the second embodiment of the present invention;

[0053] Figure 4 is a schematic diagram of a COT reasoning mechanism provided by a second embodiment of the present invention;

[0054] Figure 5 It is a schematic diagram of the structure of a terminal device provided in the third embodiment of the present invention. DETAILED DESCRIPTION

[0055] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0056] In order to illustrate the technical solution of the present invention, a specific embodiment is provided below for illustration.

[0057] Embodiment 1

[0058] See also Figure 1 , is a flow chart of a large-model-based exercise situation impact analysis method provided in a first embodiment of the present invention. The large-model-based exercise situation impact analysis method can be applied to any device or system. The large-model-based exercise situation impact analysis method includes the following steps:

[0059] Step S10, obtaining a drill instruction, and determining a target station and a target closed area according to the drill instruction;

[0060] Among them, the first address identifier and the second address identifier in the drill instruction are obtained, the target station is determined according to the first address identifier, and the target closed area is determined according to the second address identifier. The target closed area is the area that the user needs to close.

[0061] Step S20, obtaining a regional image of each station area in the target station, and determining crowd flow distribution information according to the regional image;

[0062] The cameras in each station area are controlled to collect images to obtain regional images, and face recognition is performed on the regional images to obtain the current number of users in each station area. Crowd distribution information is generated based on the current number of users in each station area.

[0063] Step S30, obtaining ticket sales information within a preset time period at the target station, and determining a change in passenger flow trend according to the ticket sales information;

[0064] Among them, the preset duration can be set according to needs. For example, the preset duration can be set to a time range within 10 minutes before the current time. The ticket information includes the ticket time and the user's destination. Based on the user's destination, the number of passengers in each station area in the future time can be effectively determined. Based on the number of passengers in each station area in the future time, the change in the passenger flow trend of each station area can be effectively determined.

[0065] Step S40, determining drill external information according to the station location and current time of the target station;

[0066] Among them, the drill external information includes drill weather information, drill holiday information and drill event information. Weather information is obtained according to the station location of the target station to obtain drill weather information. Whether it is a holiday is determined according to the current time to obtain drill holiday information. The surrounding event information is obtained according to the station location of the target station and the current time to obtain drill event information.

[0067] Step S50, inputting the target closed area, the crowd distribution information, the crowd trend change and the drill external information into the pre-trained large model to perform situation impact analysis, and obtain the drill situation impact analysis result;

[0068] Among them, the use of a large model can automatically conduct an exercise situation impact analysis on the target closed area based on crowd distribution information, crowd trend changes, exercise weather information, exercise holiday information and exercise event information. There is no need to use manual experience to conduct exercise situation impact analysis, which improves the comprehensiveness and accuracy of the exercise situation impact analysis.

[0069] Optionally, before inputting the target closed area, the crowd flow distribution information, the crowd flow trend change and the drill external information into the pre-trained large model for situation impact analysis, the method further includes:

[0070] Acquire a drill sample, and input the drill sample into the large model for crowd flow analysis to obtain drill internal sample information; wherein the drill sample can be set according to demand, and the regional crowd flow trend and regional crowd flow distribution of the sample area in the station in the drill sample can be obtained by inputting the drill sample into the large model for crowd flow analysis;

[0071] According to the drill weather samples, drill holiday samples and drill event samples in the drill samples, an impact prediction is performed on the closed sample area to obtain the drill external prediction impact; wherein, feature mapping is performed on the drill weather samples, drill holiday samples and drill event samples to obtain weather sample features, holiday sample features and event sample features, feature combination is performed on the weather sample features, holiday sample features and event sample features to obtain drill external features, feature decoding is performed on the drill external features to obtain the drill external prediction impact, and the drill external prediction impact is used to represent the degree of impact of the drill weather information, drill holiday information and drill event information on passenger travel during the current drill process, and the impact degree can be often represented in a numerical form;

[0072] According to the internal sample information of the drill, the impact of the closed sample area is predicted to obtain the internal prediction impact of the drill, and the drill prediction situation is determined according to the internal prediction impact of the drill and the external prediction impact of the drill; wherein, the internal sample information of the drill is feature mapped to obtain the internal sample feature, and the internal sample feature is feature decoded to obtain the internal prediction impact, and the internal prediction impact is used to indicate the degree of influence of the crowd distribution information and the crowd trend change on the passenger travel during the current drill;

[0073] The weight values ​​of the internal prediction impact of the exercise and the external prediction impact of the exercise are calculated to obtain the prediction situation of the exercise; wherein the weight coefficients of the internal prediction impact of the exercise and the external prediction impact of the exercise in the weight value calculation can be set according to the needs;

[0074] The model loss is determined according to the internal prediction impact of the exercise, the external prediction impact of the exercise and the prediction situation of the exercise, and the parameters of the large model are updated according to the model loss to obtain the pre-trained large model.

[0075] Furthermore, the drill samples are input into the large model for crowd flow analysis to obtain drill internal sample information, including:

[0076] Performing face recognition on the regional image of the sample area in the drill sample according to the large model to obtain the number of regional faces, and determining the crowd flow distribution sample value of the sample area according to the number of regional faces; wherein, face recognition is performed on the regional image of the sample area in the drill sample to obtain the number of regional faces in each sample area, and the number of users in the sample area is determined based on the number of regional faces to obtain the crowd flow distribution sample value;

[0077] Obtaining the ticketing sample information in the drill sample, and determining the target platform according to the user destination in the ticketing sample information; wherein the user destination is matched with a platform query table to obtain the target platform, and the platform query table stores the correspondence between different user destinations and corresponding target platforms;

[0078] Determine the number of changes in the flow of people in the sample area according to the target station, and determine the regional flow of people trend of the sample area according to the number of changes in the flow of people; wherein, the target station is matched with the sample area, the number of successful matches in each sample area is determined as the number of target users, the difference between the number of target users and the number of regional faces is calculated to obtain the number of changes in the flow of people, and the number of changes in the flow of people is matched with the trend query table to obtain the regional flow of people trend;

[0079] Furthermore, determining the model loss according to the internal predicted impact of the exercise, the external predicted impact of the exercise and the predicted situation of the exercise includes:

[0080] Obtaining the real internal impact of the drill, the real external impact of the drill, and the real situation of the drill of the drill sample, and calculating the first loss according to the real internal impact of the drill and the predicted internal impact of the drill;

[0081] Calculating a second loss according to the actual external impact of the drill and the predicted external impact of the drill, and calculating a third loss according to the actual situation of the drill and the predicted situation of the drill;

[0082] The first loss, the second loss and the third loss are weighted to obtain the model loss; wherein, during the weight calculation process, the weight coefficients of the first loss, the second loss and the third loss can be set according to requirements.

[0083] Preferably, the target closed area, the crowd flow distribution information, the crowd flow trend change and the drill external information are input into the pre-trained large model for situation impact analysis, and after obtaining the drill situation impact analysis result, the method further includes:

[0084] Input the target closed area and the exercise external information into the pre-trained large model for exercise area matching, obtain a first candidate exercise plan, and obtain the historical exercise situation of the first candidate exercise plan; wherein, the target closed area and the exercise external information are input into the pre-trained large model for exercise area matching to obtain the real exercise plan corresponding to the target closed area under the conditions of fixed closed area and similar exercise external information, that is, the first candidate exercise plan;

[0085] The first candidate rehearsal plan corresponding to the minimum historical rehearsal situation is set as the first optimized rehearsal plan; wherein, by setting the first candidate rehearsal plan corresponding to the minimum historical rehearsal situation as the first optimized rehearsal plan, it is possible to effectively achieve the effect of minimizing the impact on passenger travel by changing the rehearsal time under the condition that the target closure area cannot be changed, and the rehearsal time in the first optimized rehearsal plan is the rehearsal optimization time.

[0086] Input the current time and the rehearsal external information into the pre-trained large model for rehearsal time matching, obtain a second candidate rehearsal plan, and obtain the historical rehearsal situation of the second candidate rehearsal plan; wherein, by inputting the current time and the rehearsal external information into the pre-trained large model for rehearsal time matching, the real rehearsal plan corresponding to the target station at the current time under the conditions of fixed rehearsal time and similar rehearsal external information, that is, the second candidate rehearsal plan, is obtained;

[0087] The second candidate rehearsal plan corresponding to the minimum historical rehearsal situation is set as the second optimized rehearsal plan, and plan optimization information is generated according to the first optimized rehearsal plan and the second optimized rehearsal plan, wherein, by setting the second candidate rehearsal plan corresponding to the minimum historical rehearsal situation as the second optimized rehearsal plan, it is possible to effectively achieve the effect of minimizing the impact on passenger travel by changing the closed area in the target station under the condition that the rehearsal time cannot be changed, and the closed area in the second optimized rehearsal plan is the rehearsal optimization area.

[0088] In this embodiment, after obtaining the drill sample, the following steps are also included:

[0089] Noise sampling is performed according to the large model to obtain sampling noise, and sample generation is performed according to the sampling noise and the training sample to obtain a generated sample; wherein a generation network and a discriminant network are provided in the large model, and the generation network is used to generate samples according to the sampling noise and the training sample to obtain a generated sample;

[0090] Performing sample judgment on the generated sample to obtain a judgment result, and determining the sample generation loss according to the judgment result; wherein the discriminant network is used to judge whether the generated sample is real data;

[0091] updating parameters of the large model according to the sample generation loss until the large model converges, and inputting the training sample into the converged large model for data enhancement to obtain enhanced data;

[0092] The enhanced data is used to replace the drill samples; wherein, by inputting the drill samples into the converged large model for data enhancement, a drill plan and situation forecast that conforms to the actual situation can be effectively generated, thereby improving the efficiency and accuracy of the drill plan data enhancement.

[0093] Specifically, the target closed area, the crowd flow distribution information, the crowd flow trend change and the drill external information are input into the pre-trained large model to perform situation impact analysis, and after obtaining the drill situation impact analysis result, the method further includes:

[0094] A drill similarity scheme is determined according to the target closed area, the crowd distribution information, the crowd trend change and the drill external information, drill problems and emergency plans of the drill similarity scheme are obtained, and drill prompts are generated according to the drill problems and the emergency plans; wherein, the target closed area, the crowd distribution information, the crowd trend change and the drill external information are matched with the drill sample to obtain data similarity, and the drill sample corresponding to the maximum data similarity is determined as the drill similarity scheme, and by generating drill prompts from the drill problems and emergency plans of the drill similarity scheme, the user can be effectively prompted with drill problems and emergency plans before the drill.

[0095] In this embodiment, a large model is adopted to conduct an exercise situation impact analysis on the target closed area based on crowd distribution information, crowd trend changes, exercise weather information, exercise holiday information and exercise event information, thereby improving the comprehensiveness and accuracy of the exercise situation impact analysis and making the exercise situation impact analysis results more accurately reflect the actual situation.

[0096] Embodiment 2

[0097] See also Figure 2 , is a structural diagram of a large model-based exercise situation impact analysis system 100 provided in a second embodiment of the present invention, including:

[0098] The area determination module 10 is used to obtain a drill instruction and determine a target station and a target closed area according to the drill instruction.

[0099] A crowd flow analysis module 11 is used to obtain a regional image of each station area in the target station and determine crowd flow distribution information according to the regional image;

[0100] The ticket sales information within a preset time period in the target station is obtained, and the change of the passenger flow trend is determined according to the ticket sales information.

[0101] The external analysis module 12 is used to determine the drill external information according to the station location and current time of the target station, wherein the drill external information includes drill weather information, drill holiday information and drill event information.

[0102] The situation analysis module 13 is used to input the target closed area, the crowd distribution information, the crowd trend change and the drill external information into the pre-trained large model to perform situation impact analysis and obtain the drill situation impact analysis result.

[0103] Preferably, the situation analysis module 13 is further used to: input the target closed area and the drill external information into the pre-trained large model to perform drill area matching, obtain a first candidate drill plan, and obtain the historical drill situation of the first candidate drill plan;

[0104] Setting the first candidate exercise plan corresponding to the minimum historical exercise situation as the first optimized exercise plan;

[0105] Input the current time and the drill external information into the pre-trained large model to match the drill time, obtain a second candidate drill plan, and obtain the historical drill situation of the second candidate drill plan;

[0106] The second candidate exercise plan corresponding to the minimum historical exercise situation is set as a second optimized exercise plan, and plan optimization information is generated according to the first optimized exercise plan and the second optimized exercise plan.

[0107] The model training module 14 is used to obtain drill samples and input the drill samples into the large model for crowd flow analysis to obtain drill internal sample information;

[0108] According to the drill weather samples, drill holiday samples and drill event samples in the drill samples, an impact prediction is performed on the closed sample area to obtain the external predicted impact of the drill;

[0109] Performing an impact prediction on the capped sample area according to the drill internal sample information to obtain the drill internal prediction impact, and determining the drill prediction situation according to the drill internal prediction impact and the drill external prediction impact;

[0110] The model loss is determined according to the internal prediction impact of the exercise, the external prediction impact of the exercise and the prediction situation of the exercise, and the parameters of the large model are updated according to the model loss to obtain the pre-trained large model.

[0111] Furthermore, the model training module 14 is also used to: obtain a drill sample, and input the drill sample into the large model for crowd flow analysis to obtain drill internal sample information;

[0112] According to the drill weather samples, drill holiday samples and drill event samples in the drill samples, an impact prediction is performed on the closed sample area to obtain the external predicted impact of the drill;

[0113] Performing an impact prediction on the capped sample area according to the drill internal sample information to obtain the drill internal prediction impact, and determining the drill prediction situation according to the drill internal prediction impact and the drill external prediction impact;

[0114] The model loss is determined according to the internal prediction impact of the exercise, the external prediction impact of the exercise and the prediction situation of the exercise, and the parameters of the large model are updated according to the model loss to obtain the pre-trained large model.

[0115] Furthermore, the model training module 14 is also used to: perform face recognition on the regional image of the sample area in the training sample according to the large model to obtain the number of regional faces, and determine the sample value of the crowd distribution of the sample area according to the number of regional faces;

[0116] Acquire the ticketing sample information in the drill sample, and determine the target platform according to the user destination in the ticketing sample information;

[0117] Determine the number of changes in the flow of people in the sample area according to the target station, and determine the regional flow of people trend in the sample area according to the number of changes in the flow of people;

[0118] The drill internal sample information includes the regional passenger flow trend and the regional passenger flow distribution.

[0119] Preferably, the model training module 14 is further used to: obtain the real internal impact of the drill, the real external impact of the drill and the real situation of the drill of the drill sample, and calculate the first loss according to the real internal impact of the drill and the predicted internal impact of the drill;

[0120] Calculating a second loss according to the actual external impact of the drill and the predicted external impact of the drill, and calculating a third loss according to the actual situation of the drill and the predicted situation of the drill;

[0121] Weight values ​​are calculated for the first loss, the second loss, and the third loss to obtain the model loss.

[0122] Specifically, the model training module 14 is further used to: perform noise sampling according to the large model to obtain sampling noise, and perform sample generation according to the sampling noise and the training sample to obtain a generated sample;

[0123] Performing sample judgment on the generated sample to obtain a judgment result, and determining the sample generation loss according to the judgment result;

[0124] updating parameters of the large model according to the sample generation loss until the large model converges, and inputting the training sample into the converged large model for data enhancement to obtain enhanced data;

[0125] The training sample is replaced by the enhanced data.

[0126] See also Figure 3 The big model-based exercise situation impact analysis system 100 uses an advanced big language model (big model) as an intelligent engine. The big language model deeply integrates the RAG (Retrieval-Augmented Generation) knowledge retrieval enhancement technology, and has undergone extensive and in-depth internal and external data training, and is optimized specifically for the specific needs of the station end-capping exercise field. The big language model has not only accumulated a wealth of professional knowledge and strategies, but also has powerful data processing and analysis capabilities, and can efficiently generate practical exercise plans. Among them, the training data involved in the pre-trained big model are as follows:

[0127] 1. Pre-training data: Pre-training data is the source of system intelligence, covering data sources in multiple dimensions. These data include but are not limited to daily station operation records, historical end-blocking drill cases, expert knowledge base, weather data, holiday schedules, information on major surrounding events and activities, and environmental monitoring data. The above data has been carefully screened, cleaned, and labeled to ensure that the model can fully and accurately absorb and understand relevant knowledge during the training process, laying a solid foundation for subsequent multimodal data fusion and analysis.

[0128] 2. Multimodal data fusion and understanding: The system uses advanced data fusion technology to achieve deep integration and understanding of multi-source heterogeneous data. Specifically, the system can acquire and process the crowd flow video monitored by cameras, the crowd flow transaction data of automatic ticket vending machines, and environmental monitoring data in real time, while combining multi-dimensional information such as weather, holidays, and major surrounding events and activities. These data are converted into structured and standardized data formats through complex processing processes such as video image recognition, natural language processing, and time series analysis, providing comprehensive and accurate data support for subsequent situation analysis and simulation.

[0129] 3. The system combines retrieval technology and generation technology: In the exercise plan generation and situation simulation stage, the system gives full play to the advantages of RAG technology. On the one hand, through the powerful retrieval function, the system can quickly retrieve information and strategy fragments that are highly relevant to the current exercise scenario from the historical case library and expert knowledge base; on the other hand, using generation technology, the system can dynamically and creatively generate exercise plans and situation forecasts that conform to the actual situation based on these retrieval results and the current multimodal data input. Not only does it improve the efficiency and accuracy of plan generation, but it also enables the system to flexibly respond to various new exercise scenarios and challenges.

[0130] During the pre-training process of the large language model, human feedback reinforcement learning RLHF (Reinforcement Learning from Human Feedback) is performed based on the pre-training data, and the large language model after human feedback reinforcement learning is fine-tuned to obtain the pre-trained large language model.

[0131] See also Figure 4 The system integrates the advanced pre-trained large language model and Chain-of-Thought (COT) reasoning mechanism. By building an intelligent simulation platform and integrating multi-dimensional data resources of station end-capping drills, it forms an intelligent system that can efficiently predict and evaluate the impact of drills.

[0132] Core engine and reasoning mechanism: The system uses a pre-trained large language model as the core intelligent engine. The model has been trained with a large amount of station operation and drill data, and is well aware of the complex scenarios and potential impacts of station end-blocking drills. Combined with the COT reasoning mechanism, the system can simulate the human thinking process and conduct logically rigorous analysis and prediction of complex drill situations. Specific application cases are as follows:

[0133] Initial conditions: The station receives an end-sealing drill instruction and must immediately close one end of the station to simulate the emergency response process.

[0134] COT Reasoning 1: Comprehensive Evaluation of Internal and External Data

[0135] Reasoning process: First, use camera data to analyze the current passenger flow distribution in various areas of the station in real time. At the same time, combine the passenger flow data of the automatic vending machine to determine the passenger flow trend and the areas that may be affected.

[0136] Next, consider external factors such as weather conditions (e.g., rainy days may cause passengers to be stranded), holidays (e.g., holiday traffic may increase significantly), and major surrounding events and activities (e.g., a large influx of audiences after a concert), and evaluate their impact on the execution of the drill.

[0137] At the same time, we reviewed historical drill data and analyzed problems and solutions in similar scenarios to provide reference for this drill.

[0138] Action branch: Adjust the drill plan based on the comprehensive assessment results, such as adding manpower, adjusting the blockade area, and launching broadcast notifications in advance.

[0139] COT Reasoning 2: Execution of End Capping Operations and Real-time Monitoring

[0140] Reasoning process: When performing the sealing operation, continue to use the camera and automatic vending machine data to monitor the changes in the flow of people in real time to ensure that the sealed area is effectively controlled and does not affect the normal operation of other areas.

[0141] At the same time, pay close attention to weather changes and external events, and be prepared to deal with emergencies.

[0142] Action branch: Based on real-time monitoring data, timely adjust blockade measures, such as adding cordons and strengthening traffic control.

[0143] COT Reasoning 3: Dealing with emergencies and flexible adjustments

[0144] Reasoning process: During the drill, if an emergency situation occurs (such as a sudden increase in traffic, equipment failure, etc.), the scope and extent of the impact will be comprehensively determined based on internal and external data.

[0145] For example, if camera data shows a surge in foot traffic in a certain area and a sudden increase in the transaction volume of automatic vending machines, it may mean that a large number of passengers are moving in that direction. At this time, it is necessary to analyze the reasons based on weather and surrounding activity information and quickly formulate response measures.

[0146] Action branch: Based on the analysis and judgment results, quickly adjust the drill plan or activate the emergency plan to ensure that the drill proceeds smoothly and does not affect passenger safety.

[0147] COT Reasoning 4: Exercise Summary and Data Feedback

[0148] Reasoning process: After the drill, collect and analyze data records from cameras, automatic vending machines and other equipment, as well as information from external factors such as weather, holidays, and surrounding activities to evaluate the impact on the drill results.

[0149] At the same time, based on historical drill data and the actual situation of this drill, we summarize the experience and lessons learned and provide improvement suggestions for future drills.

[0150] Action branch: Prepare detailed drill reports and data analysis reports, and submit them to relevant departments and personnel for review; organize relevant personnel to discuss and study, and improve emergency response capabilities and comprehensive data analysis capabilities.

[0151] In this embodiment, the system integrates a comprehensive knowledge base of station end-blocking drills, including but not limited to historical drill data, station operation data, traffic flow patterns, holiday impacts, etc., to provide solid data support for simulation analysis. By comprehensively utilizing data resources, the system can more accurately simulate and predict the actual impact of drills on station operations.

[0152] The system can automatically collect and analyze relevant data, use COT technology to perform multi-level and multi-dimensional logical reasoning, and achieve accurate prediction of key indicators such as changes in station passenger flow, traffic congestion, and passenger travel time delays during the drill, greatly improving the efficiency and accuracy of simulation analysis.

[0153] Based on the simulation prediction results, the system can provide station managers with scientific drill schedule arrangements, location selection, and content design suggestions, striving to minimize the impact on passengers' daily travel while ensuring the effectiveness of the drill.

[0154] The system has a strong self-learning ability, and can continuously absorb new drill cases and feedback data to optimize simulation models and prediction algorithms. With the accumulation of time and experience, the system's prediction accuracy and efficiency will continue to improve, providing strong support for the continuous optimization of station end-sealing drills.

[0155] In this embodiment, by introducing a pre-trained large language model and combining it with powerful data processing capabilities, it is possible to fully integrate multi-dimensional data resources such as weather conditions, traffic conditions, holiday information, historical site traffic, historical exercise data, etc. This all-round data integration ensures the comprehensiveness and accuracy of the analysis results, so that the prediction results can more accurately reflect the actual situation.

[0156] By using the generalization ability of the large model and the COT reasoning mechanism, multi-level and multi-dimensional logical reasoning is performed on complex drill scenarios, significantly improving the accuracy of predictions. By learning and analyzing a large amount of historical data, the system can accurately predict the potential impact of drills on station operations, passenger travel, etc., providing a scientific basis for the reasonable selection of drill time and location.

[0157] The intelligent simulation platform built is highly flexible and scalable. The system can monitor various changes in the exercise process in real time, such as sudden weather changes and traffic congestion, and quickly adjust the simulation model according to the real-time situation to ensure the timeliness and accuracy of the prediction results. At the same time, the system also supports dynamic adjustment of the exercise plan to deal with unforeseen situations.

[0158] More importantly, the system in this embodiment has the ability to continuously learn and can continuously absorb new fault cases and solutions over time, thereby continuously improving its own diagnostic accuracy and fault handling efficiency, effectively overcoming the problem of knowledge loss caused by personnel mobility.

[0159] Through multi-channel information push mechanisms (such as mobile APP notifications, social media releases, station large screens, etc.), the drill information can be widely covered and notified instantly. This can not only improve the timeliness and comprehensiveness of information transmission, but also enhance passengers' understanding and cooperation with the drill, and reduce misunderstandings and dissatisfaction.

[0160] By collecting and analyzing passengers’ feedback on the drill, the system can timely understand the impact of the drill on passengers’ travel and identify existing problems and deficiencies. Based on these feedback data, the system can continuously optimize the simulation model and prediction algorithm, improve the scientificity and rationality of the drill, and achieve continuous improvement and optimization.

[0161] Embodiment 3

[0162] Figure 5 2 is a block diagram of a terminal device 2 provided in the third embodiment of the present application. Figure 5 As shown, the terminal device 2 of this embodiment includes: a processor 20, a memory 21, and a computer program 22 stored in the memory 21 and executable on the processor 20, such as a program of a method for analyzing the impact of a drill situation based on a large model. When the processor 20 executes the computer program 22, the steps in each embodiment of the method for analyzing the impact of a drill situation based on a large model are implemented.

[0163] Exemplarily, the computer program 22 may be divided into one or more modules, which are stored in the memory 21 and executed by the processor 20 to complete the present application. The one or more modules may be a series of computer program instruction segments capable of completing specific functions, which are used to describe the execution process of the computer program 22 in the terminal device 2. The terminal device may include, but is not limited to, a processor 20 and a memory 21.

[0164] The processor 20 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor, etc.

[0165] The memory 21 may be an internal storage unit of the terminal device 2, such as a hard disk or memory of the terminal device 2. The memory 21 may also be an external storage device of the terminal device 2, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. equipped on the terminal device 2. Further, the memory 21 may also include both an internal storage unit and an external storage device of the terminal device 2. The memory 21 is used to store the computer program and other programs and data required by the terminal device. The memory 21 may also be used to temporarily store data that has been output or is to be output.

[0166] In addition, each functional module in each embodiment of the present application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The above integrated unit can be implemented in the form of hardware or in the form of software functional unit.

[0167] If the integrated module is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Among them, the computer-readable storage medium can be non-volatile or volatile. Based on this understanding, the present application implements all or part of the processes in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium, and the computer program can implement the steps of the above-mentioned various method embodiments when executed by the processor. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable storage medium may include: any entity or device capable of carrying computer program code, recording medium, U disk, mobile hard disk, disk, optical disk, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electric carrier signal, telecommunication signal and software distribution medium, etc. It should be noted that the content contained in computer-readable storage media can be appropriately increased or decreased according to the requirements of legislation and patent practices in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practices, computer-readable storage media do not include electrical carrier signals and telecommunication signals.

[0168] The embodiments described above are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, a person skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application, and should all be included in the protection scope of the present application.

Claims

1. A method for analyzing the impact of exercise situation based on a large model, characterized in that: The method comprises: Obtaining a drill instruction, and determining a target station and a target blocking area according to the drill instruction; Acquire a regional image of each station area in the target station, and determine crowd flow distribution information according to the regional image; Acquire ticket sales information within a preset time period at the target station, and determine changes in passenger flow trends based on the ticket sales information; Determining drill external information according to the station location and current time of the target station, the drill external information including drill weather information, drill holiday information and drill event information; The target closed area, the crowd distribution information, the crowd trend change and the drill external information are input into the pre-trained large model to perform situation impact analysis to obtain the drill situation impact analysis result.

2. The exercise situation impact analysis method based on a large model as claimed in claim 1, characterized in that: Before inputting the target closed area, the crowd flow distribution information, the crowd flow trend change and the drill external information into the pre-trained large model for situation impact analysis, the method further includes: Acquire drill samples, and input the drill samples into the large model for crowd flow analysis to obtain drill internal sample information; According to the drill weather samples, drill holiday samples and drill event samples in the drill samples, an impact prediction is performed on the closed sample area to obtain the external predicted impact of the drill; Performing an impact prediction on the capped sample area according to the drill internal sample information to obtain the drill internal prediction impact, and determining the drill prediction situation according to the drill internal prediction impact and the drill external prediction impact; The model loss is determined according to the internal prediction impact of the exercise, the external prediction impact of the exercise and the prediction situation of the exercise, and the parameters of the large model are updated according to the model loss to obtain the pre-trained large model.

3. The exercise situation impact analysis method based on a large model as claimed in claim 2 is characterized in that: The drill samples are input into the large model for crowd flow analysis to obtain drill internal sample information, including: Performing face recognition on the regional image of the sample area in the drill sample according to the large model to obtain the number of regional faces, and determining the crowd flow distribution sample value of the sample area according to the number of regional faces; Acquire the ticketing sample information in the drill sample, and determine the target platform according to the user destination in the ticketing sample information; Determine the number of changes in the flow of people in the sample area according to the target station, and determine the regional flow of people trend in the sample area according to the number of changes in the flow of people; The drill internal sample information includes the regional passenger flow trend and the regional passenger flow distribution.

4. The exercise situation impact analysis method based on a large model as claimed in claim 2 is characterized in that: Determining the model loss according to the internal predicted impact of the exercise, the external predicted impact of the exercise, and the predicted situation of the exercise includes: Obtaining the real internal impact of the drill, the real external impact of the drill, and the real situation of the drill of the drill sample, and calculating the first loss according to the real internal impact of the drill and the predicted internal impact of the drill; Calculating a second loss according to the actual external impact of the drill and the predicted external impact of the drill, and calculating a third loss according to the actual situation of the drill and the predicted situation of the drill; Weight values ​​are calculated for the first loss, the second loss, and the third loss to obtain the model loss.

5. The large model-based exercise situation impact analysis method according to claim 1, characterized in that: After inputting the target closed area, the crowd flow distribution information, the crowd flow trend change and the drill external information into the pre-trained large model for situation impact analysis and obtaining the drill situation impact analysis result, the method further includes: Input the target closed area and the drill external information into the pre-trained large model to perform drill area matching, obtain a first candidate drill plan, and obtain the historical drill situation of the first candidate drill plan; Setting the first candidate exercise plan corresponding to the minimum historical exercise situation as the first optimized exercise plan; Input the current time and the drill external information into the pre-trained large model to match the drill time, obtain a second candidate drill plan, and obtain the historical drill situation of the second candidate drill plan; The second candidate exercise plan corresponding to the minimum historical exercise situation is set as a second optimized exercise plan, and plan optimization information is generated according to the first optimized exercise plan and the second optimized exercise plan.

6. The exercise situation impact analysis method based on a large model as claimed in claim 2, characterized in that: After getting the walkthrough sample, also included: Noise sampling is performed according to the large model to obtain sampling noise, and sample generation is performed according to the sampling noise and the training sample to obtain a generated sample; Performing sample judgment on the generated sample to obtain a judgment result, and determining the sample generation loss according to the judgment result; updating parameters of the large model according to the sample generation loss until the large model converges, and inputting the training sample into the converged large model for data enhancement to obtain enhanced data; The training sample is replaced by the enhanced data.

7. The large model-based exercise situation impact analysis method according to claim 1, characterized in that: After inputting the target closed area, the crowd flow distribution information, the crowd flow trend change and the drill external information into the pre-trained large model for situation impact analysis and obtaining the drill situation impact analysis result, the method further includes: Determine a drill similarity scheme according to the target closed area, the crowd flow distribution information, the crowd flow trend change and the drill external information, and obtain drill problems and emergency plans for the drill similarity scheme; A drill prompt is generated based on the drill question and the emergency plan.

8. A large-model-based exercise situation impact analysis system, characterized in that: The system comprises: An area determination module is used to obtain a drill instruction and determine a target station and a target end-blocking area according to the drill instruction; A crowd flow analysis module, used to obtain a regional image of each station area in the target station, and determine crowd flow distribution information according to the regional image; Acquire ticket sales information within a preset time period at the target station, and determine changes in passenger flow trends based on the ticket sales information; An external analysis module, used to determine drill external information according to the station location and current time of the target station, wherein the drill external information includes drill weather information, drill holiday information and drill event information; The situation analysis module is used to input the target closed area, the crowd distribution information, the crowd trend change and the drill external information into the pre-trained large model to perform situation impact analysis and obtain the drill situation impact analysis results.

9. The large model-based exercise situation impact analysis system according to claim 8, characterized in that: The exercise situation impact analysis system also includes: The model training module is used to obtain drill samples and input the drill samples into the large model for crowd flow analysis to obtain drill internal sample information; According to the drill weather samples, drill holiday samples and drill event samples in the drill samples, an impact prediction is performed on the closed sample area to obtain the external predicted impact of the drill; Performing an impact prediction on the capped sample area according to the drill internal sample information to obtain the drill internal prediction impact, and determining the drill prediction situation according to the drill internal prediction impact and the drill external prediction impact; The model loss is determined according to the internal prediction impact of the exercise, the external prediction impact of the exercise and the prediction situation of the exercise, and the parameters of the large model are updated according to the model loss to obtain the pre-trained large model.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images