Emergency Management System and Method for Urban Rail Transit Depots Based on AI Big Data Model

Abstract

This invention discloses an emergency management system and method for urban rail transit depots based on an AI large-scale model, belonging to the field of urban rail transit depot emergency management technology. It includes: a multi-source data acquisition and fusion processing module for acquiring multi-dimensional detection data related to emergency events within the depot and performing weighted fusion; an AI large-scale model training and optimization module for using the fused data as training samples and inputting them into a pre-built AI large-scale model, training the AI ​​large-scale model by dynamically adjusting the learning rate; an emergency event intelligent identification and early warning module for identifying and issuing early warnings for emergency events based on real-time acquired multi-dimensional detection data and the AI ​​large-scale model; and an intelligent response plan generation module for generating the optimal response plan based on the comprehensive risk value of the emergency event. This invention achieves accurate identification, processing, and assessment of emergency events in the depot, efficient resource allocation, and improves the efficiency and effectiveness of emergency response.

Description

Technical Field

[0001] This invention relates to the field of emergency management technology for urban rail transit depots, specifically to an emergency management system and method for urban rail transit depots based on an AI big data model. Background Technology

[0002] With the rapid development of urban rail transit, depots have become a crucial link in ensuring the safe, stable, and efficient operation of the urban rail transit network. Especially in emergency situations, the efficiency of depot emergency response directly affects the efficiency of urban rail transit and even the entire city's operation. Traditionally, emergency response management in urban rail transit depots relies heavily on manual operation. Typically, before an emergency occurs, procedures and measures for handling different emergency situations are pre-established. When an emergency occurs, depot maintenance personnel only need to handle the emergency according to the pre-established procedures and measures.

[0003] This traditional emergency response model has several shortcomings in practical application. First, it lacks effective means to integrate multi-dimensional emergency monitoring data, making it difficult for managers to quickly obtain comprehensive and accurate on-site information and to comprehensively assess the severity and development trend of emergencies based on multi-dimensional emergency monitoring data. Second, it lacks data analysis and intelligent identification methods for the integrated emergency monitoring data, making it difficult to issue timely warnings in the early stages of an event, often missing the optimal response opportunity. Third, the traditional operation and maintenance model relies too heavily on pre-defined fixed procedures, lacking flexibility and real-time adaptability. It struggles to quickly adjust response plans based on actual on-site conditions, and managers often develop rigid thinking in complex and ever-changing emergency scenarios, failing to make optimal decisions for the current situation. Furthermore, the traditional operation and maintenance model lacks effective means to comprehensively measure response time, resource utilization efficiency, and emergency resource costs during emergency response. Finally, the traditional system lacks a continuous optimization mechanism, unable to automatically learn and optimize response plans from historical emergency events. After each emergency response, the lack of an effective feedback mechanism makes it difficult to improve future response strategies.

[0004] Therefore, there is an urgent need for a comprehensive and intelligent emergency response solution for the field that can effectively improve the efficiency and effectiveness of emergency response in the field. Summary of the Invention

[0005] The technical problem this invention aims to solve is the slow response speed, lack of scientific basis for decision-making, and low efficiency of resource allocation in existing urban rail transit depot emergency management systems. The purpose of this invention is to provide an urban rail transit depot emergency management system and method based on an AI-powered large-scale model, enabling accurate identification of depot emergency events, intelligent decision support, efficient resource allocation, and the establishment of a continuous optimization mechanism to improve the efficiency and effectiveness of emergency response.

[0006] This invention is achieved through the following technical solution:

[0007] In a first aspect, the present invention provides an emergency management system for urban rail transit depots based on an AI large-scale model, the system comprising:

[0008] The multi-source data acquisition and fusion processing module is used to acquire multi-dimensional detection data related to emergency events within the field, and to weight and fuse the multi-dimensional detection data with relevant factors to obtain the fused data;

[0009] The AI ​​large model training and optimization module is used to input the fused data, historical emergency event case data, expert experience data, and field emergency event management data as training samples into the pre-built AI large model. The AI ​​large model is trained by dynamically adjusting the learning rate to obtain a trained AI large model.

[0010] The emergency event intelligent identification and early warning module is used to extract key feature information related to emergency events based on real-time multi-dimensional detection data, input the key feature information into a pre-trained AI model, identify emergency events based on the pre-trained AI model, and evaluate the comprehensive risk value of emergency events based on the identification results, and issue early warnings for emergency situations that exceed the risk threshold.

[0011] The intelligent response plan generation module is used to generate multiple response plans based on the comprehensive risk value of the emergency event and a trained AI model, and to conduct a comprehensive evaluation of the response plans to generate the optimal response plan.

[0012] Furthermore, the system also includes:

[0013] The emergency response assessment and model optimization module is used to evaluate the effectiveness of emergency response based on relevant data during the emergency process; and to feed the assessment results and emergency response process data back to the AI ​​big model for model optimization and updates.

[0014] Furthermore, the multi-source data acquisition and fusion processing module includes:

[0015] The multi-source data acquisition unit is used to classify emergency events according to the pre-sorted types of emergency events in the field, and to collect multi-dimensional detection data related to each type of emergency event in real time; for example, for fire emergency events in the field, the multi-dimensional detection data may include temperature data, smoke concentration data and video data of the fire accident.

[0016] The fusion processing unit is used to preprocess the multi-dimensional detection data to obtain preprocessed feature parameters; and to perform weighted fusion of the preprocessed feature parameters with relevant factors, and to incorporate the time factor in the fusion process to obtain fused data.

[0017] Furthermore, the expression for weighted fusion is:

[0018] ;

[0019] in, The merged data; Indicates emergency characteristic parameters At any moment Real-time value; , Emergency characteristic parameters The baseline value and the upper limit of the threshold; Emergency characteristic parameters The risk assessment value; for The maximum value; , , These are the weighting coefficient, risk enhancement coefficient, and time decay coefficient, respectively. and These are the duration of the emergency event and the maximum duration that the site can withstand for the emergency event, respectively. These are the data dimensions involved in emergency events.

[0020] Furthermore, the AI ​​large model training and optimization module includes:

[0021] The AI ​​large model training unit is used to input the fused data, historical emergency event case data, expert experience data, and field emergency event management data as training samples into the pre-built AI large model. The AI ​​large model is trained by dynamically adjusting the learning rate to obtain a trained AI large model.

[0022] The optimization unit is used to continuously evaluate and validate the model performance using the validation dataset during the training of large AI models, and to adjust and optimize the model parameters based on the evaluation results; it also introduces a meta-learning strategy to adapt to new scenarios and data changes.

[0023] Furthermore, dynamically adjusting the learning rate introduces hyperparameters during the training of large AI models. And construct the learning rate dynamic adjustment coefficient based on the loss function values ​​before and after training. The expression for dynamically adjusting the learning rate is:

[0024] ;

[0025] ;

[0026] in, This represents the current learning rate. This indicates the learning efficiency of the previous training session. Indicates the current training loss. This indicates the loss from the previous training session. This represents the dynamic adjustment coefficient for the current training. This represents the dynamic adjustment coefficient for the current training iteration. It is used for control Adjust the speed hyperparameter, with a value between 0 and 1.

[0027] Furthermore, based on the trained AI model, emergency events are identified; and combined with the identification results, the comprehensive risk value of emergency events is assessed, and early warnings are issued for emergency situations exceeding the risk threshold, including:

[0028] The system uses a pre-trained AI model to identify emergency events and obtain identification results. The identification results include the type, severity, scope of impact, and development trend of the emergency event.

[0029] Based on the identification results, a comprehensive risk calculation formula is used to assess the comprehensive risk value of the emergency event, resulting in the comprehensive risk value of the emergency event; the comprehensive risk calculation formula is as follows: In the formula, This represents the overall risk value of an emergency event; Indicates the type of emergency event The severity level, ranging from 1 to 5; Indicates the type of emergency event The range of influence, with values ​​ranging from 1 to 5; Indicates the type of emergency event The probability of occurrence ranges from 0 to 1; This indicates the number of identified emergency event types;

[0030] The overall risk value of an emergency event is compared with a preset risk threshold. When the overall risk value of an emergency event exceeds the preset risk threshold, an early warning signal is issued.

[0031] Furthermore, the expression for the comprehensive evaluation of the disposal plan is as follows:

[0032] ;

[0033] in, This indicates the overall evaluation score of the disposal plan. This represents the score of the treatment plan on the effectiveness dimension k (value range 0 to 100). Indicates the weight of the effect dimension k (value range from 0 to 1); This represents the score of the solution on the risk dimension l (with a value range of 0 to 100). This represents the weight of risk dimension l (ranging from 0 to 1). ; Indicates the number of effect dimensions. This indicates the number of risk dimensions.

[0034] Furthermore, the expression for evaluating the effectiveness of emergency response is as follows:

[0035] ;

[0036] in, This indicates the score for evaluating the effectiveness of the emergency response; Indicates the emergency response time; This indicates the maximum emergency response time according to the emergency management regulations for the site. Indicates resource utilization efficiency; This indicates the maximum resource utilization rate according to the emergency management regulations for the site. This indicates the score for evaluating the effectiveness of the emergency response; This indicates the maximum emergency response score based on the emergency management regulations for the site. This indicates the total cost of emergency resources; This indicates the maximum total cost of emergency resources according to the emergency management regulations for the site. , , , These represent the weighting coefficients for emergency response time, resource utilization rate, emergency effectiveness evaluation score, and total emergency resource cost, respectively. .

[0037] Secondly, this invention also provides an emergency management method for urban rail transit depots based on an AI large-scale model, the method comprising:

[0038] Acquire multi-dimensional detection data related to emergency events within the field area, and then weight and fuse the multi-dimensional detection data with relevant factors to obtain the fused data;

[0039] The fused data, historical emergency event case data, expert experience data, and field emergency event management data are used as training samples and input into the pre-built AI big model. The AI ​​big model is trained by dynamically adjusting the learning rate to obtain a trained AI big model.

[0040] Based on real-time acquired multi-dimensional detection data, key feature information related to emergency events is extracted and input into a pre-trained AI model. The pre-trained AI model is then used to identify emergency events. The comprehensive risk value of emergency events is assessed based on the identification results, and warnings are issued for emergency situations that exceed the risk threshold.

[0041] Based on the comprehensive risk value of the emergency, multiple response plans are generated using a pre-trained AI model, and a comprehensive evaluation of these plans is conducted to generate the optimal response plan.

[0042] Furthermore, the method also includes:

[0043] Based on relevant data during the emergency response, an evaluation of the emergency response effectiveness is conducted; and the evaluation results and emergency response process data are fed back to the AI ​​big model for model optimization and updates.

[0044] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0045] 1. This invention relates to an emergency management system and method for urban rail transit depots based on an AI large model, which enables accurate identification, processing and assessment of emergency events in depots, intelligent decision support and efficient resource allocation, and establishes a continuous optimization mechanism to improve the efficiency and effectiveness of emergency response.

[0046] 2. The multi-source data acquisition and fusion processing module of the present invention integrates multi-dimensional detection data, taking into account factors such as data benchmark, threshold upper limit, weight influence, risk enhancement influence, time decay factor, duration of emergency events and maximum duration that the field can withstand, etc., and uses weighted fusion to construct a comprehensive risk assessment index, thereby realizing a comprehensive quantitative assessment of risk data and providing reliable data support for subsequent decision-making.

[0047] 3. The AI ​​large model training and optimization module of this invention introduces a dynamic adjustment strategy for hyperparameters and learning rate, which automatically adjusts the learning rate according to the change of loss function value, thereby improving model training efficiency and performance and ensuring stable model convergence.

[0048] 4. The emergency event intelligent identification and early warning module of the present invention uses an AI big data model to accurately identify the type and severity of emergency events, and combines the probability of occurrence to assess the comprehensive risk value and issue early warning signals in a timely manner.

[0049] 5. The intelligent solution generation module of this invention selects suitable solutions from historical cases and expert experience, and uses a comprehensive solution evaluation formula to quantitatively evaluate each solution from the dimensions of effectiveness and risk, assisting decision-makers in quickly selecting the optimal solution.

[0050] 6. The emergency process assessment and model optimization module of this invention collects emergency process data, quantitatively assesses the effectiveness of emergency response, and feeds the assessment results back to the AI ​​big model for optimization training. It continuously optimizes the model by combining newly collected case data, forming a closed-loop optimization mechanism to continuously improve the model and system functions and enhance the system's adaptability and reliability. Attached Figure Description

[0051] The accompanying drawings, which are included to provide a further understanding of embodiments of the invention and form part of this application, do not constitute a limitation thereof. In the drawings:

[0052] Figure 1 This is a structural block diagram of the urban rail transit depot emergency management system based on an AI large model, as presented in this invention.

[0053] Figure 2 This is a flowchart of the multi-source data acquisition and fusion processing module of the present invention;

[0054] Figure 3 This is a flowchart of the AI ​​large model training and optimization module of the present invention;

[0055] Figure 4 This is a flowchart of the emergency event intelligent identification and early warning module of the present invention;

[0056] Figure 5 This is a flowchart of the intelligent generation module for the treatment scheme of the present invention;

[0057] Figure 6 This is a flowchart of the emergency process assessment and model optimization module of the present invention;

[0058] Figure 7 This is a flowchart of the emergency management method for urban rail transit depots based on an AI large model, as described in this invention. Detailed Implementation

[0059] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the embodiments and accompanying drawings. The illustrative embodiments and descriptions of the present invention are only used to explain the present invention and are not intended to limit the present invention.

[0060] The key technical points of this invention are: the urban rail transit depot emergency management system based on an AI large-scale model, including a multi-source data acquisition and fusion processing module, an AI large-scale model training and optimization module, an emergency event intelligent identification and early warning module, a response plan intelligent generation module, and an emergency process evaluation and model optimization module. Specifically, as follows:

[0061] 1. Multi-source data acquisition and fusion processing module: This module integrates multi-dimensional detection data related to emergency events. It integrates the monitored characteristic data indicators with their benchmarks, upper thresholds, weights, risk enhancement, time decay, and factors such as the duration of the emergency event and the maximum duration that the site can withstand. By weighting and integrating each parameter and incorporating the time factor, the integrated data is obtained, which is the comprehensive risk assessment index for emergency events, thereby achieving a quantitative assessment of the risks of emergency events.

[0062] 2. AI large model training and optimization module, which introduces hyperparameters during model training. And construct the learning rate dynamic adjustment coefficient based on the loss function values ​​before and after training. ; By calculating the current loss Compared to the previous loss Relative changes, dynamic adjustments If the current loss is relatively small, it indicates that training is progressing smoothly. The dynamic adjustment coefficient is increased, prompting the learning rate to rise and accelerating training. Conversely, if the current loss is relatively large, the dynamic adjustment coefficient is decreased, lowering the learning rate to ensure training stability. The updated dynamic adjustment coefficient is then used to adjust the current learning rate. The learning rate at the previous moment Based on this, the learning rate is finely adjusted according to the rate of change of the loss. If the loss decreases, the learning rate is increased appropriately to accelerate convergence; if the loss fluctuates or increases, the learning rate is decreased to prevent training divergence. This strategy enables the model to automatically adjust the learning rate according to the training status, improving training efficiency and model performance.

[0063] 3. Emergency Event Intelligent Identification and Early Warning Module: By acquiring monitoring data of emergency events, it uses AI big data models to accurately identify emergency events, assesses the comprehensive risk value of emergency events based on their severity, scope of impact, and probability of occurrence, and issues early warnings for emergency situations that exceed the risk threshold.

[0064] 4. The intelligent response plan generation module, upon receiving the comprehensive risk value from the emergency event intelligent identification and early warning module, selects suitable solutions from a massive database of historical emergency cases and expert experience. Its core lies in using a comprehensive evaluation formula for response plans to quantitatively assess each plan from both effectiveness (e.g., recovery time, reduction of property damage) and risk (e.g., risk of casualties, risk of fire spread) dimensions, deriving a comprehensive evaluation score for each plan. This not only provides decision-makers with multiple targeted and actionable intelligent response plan options but also, through quantitative analysis using the formula, assists them in quickly and accurately selecting the optimal plan, significantly improving the scientific rigor and accuracy of emergency response decisions and ensuring the rational allocation of emergency resources and maximizing response efficiency.

[0065] 5. The Emergency Process Assessment and Model Optimization module collects and organizes various data from the emergency process (such as emergency response time, resource allocation efficiency, and personnel performance) after the emergency is handled. Using emergency response effectiveness evaluation formulas, it quantitatively assesses the overall effectiveness of emergency response work from multiple dimensions, including emergency response time, resource utilization rate, emergency effectiveness evaluation score, and total emergency resource cost. The evaluation results not only provide intuitive performance feedback for emergency management departments but also serve as an important reference for optimizing and training the AI ​​large-scale model. Combined with newly collected emergency case data, the AI ​​large-scale model undergoes continuous incremental learning and optimization training, constantly improving the model's understanding and perception of various emergency scenarios in urban rail transit sections, and enhancing the model's performance in emergency scenario identification, decision-making suggestions, and resource allocation.

[0066] Example 1

[0067] like Figure 1 As shown, this invention relates to an urban rail transit depot emergency management system based on an AI large-scale model. The system includes:

[0068] The multi-source data acquisition and fusion processing module is used to acquire multi-dimensional detection data related to emergency events within the field, and to weight and fuse the multi-dimensional detection data with relevant factors to obtain the fused data;

[0069] The AI ​​large model training and optimization module is used to input the fused data, historical emergency event case data, expert experience data, and field emergency event management data as training samples into the pre-built AI large model. The AI ​​large model is trained by dynamically adjusting the learning rate to obtain a trained AI large model.

[0070] The emergency event intelligent identification and early warning module is used to extract key feature information related to emergency events based on real-time multi-dimensional detection data, input the key feature information into a pre-trained AI model, identify emergency events based on the pre-trained AI model, and evaluate the comprehensive risk value of emergency events based on the identification results, and issue early warnings for emergency situations that exceed the risk threshold.

[0071] The intelligent response plan generation module is used to generate multiple response plans based on the comprehensive risk value of the emergency event and a trained AI model, and to conduct a comprehensive evaluation of the response plans to generate the optimal response plan.

[0072] The emergency response assessment and model optimization module is used to evaluate the effectiveness of emergency response based on relevant data during the emergency process; and to feed the assessment results and emergency response process data back to the AI ​​big model for model optimization and updates.

[0073] In this embodiment, the multi-source data acquisition and fusion processing module includes:

[0074] The multi-source data acquisition unit is used to classify emergency events according to the pre-sorted types of emergency events in the field, and to collect multi-dimensional detection data related to each emergency event using various sensors and monitoring equipment, including temperature data, smoke concentration data and video data of fire accidents.

[0075] The fusion processing unit is used to preprocess multi-dimensional detection data (including data cleaning and transformation), remove redundancy and noise, unify data format and semantics, and obtain preprocessed feature parameters; and then weight and fuse the preprocessed feature parameters with relevant factors, and incorporate time factors in the fusion process to obtain fused data.

[0076] The flowchart of the multi-source data acquisition and fusion processing module is as follows: Figure 2 As shown.

[0077] Specifically, the expression for weighted fusion is:

[0078] (1);

[0079] in, The merged data; Indicates emergency characteristic parameters At any moment Real-time value; , Emergency characteristic parameters The baseline value and the upper limit of the threshold; Emergency characteristic parameters The risk assessment value; for The maximum value; , , These are the weighting coefficient, risk enhancement coefficient, and time decay coefficient, respectively. and These are the duration of the emergency event and the maximum duration that the site can withstand for the emergency event, respectively. These are the data dimensions involved in emergency events.

[0080] The above method, by weighting and fusing various parameters and incorporating the time factor, yields the fused data, which is the comprehensive risk assessment index for emergency events. This fused data can then be stored in a structured field emergency management data knowledge base, providing high-quality data support for subsequent AI large-scale model training.

[0081] In this embodiment, the AI ​​large model training and optimization module includes:

[0082] The AI ​​large model training unit is used to input the fused data, historical emergency event case data, expert experience data, and field emergency event management data as training samples into the pre-built AI large model. The AI ​​large model is trained by dynamically adjusting the learning rate to obtain a trained AI large model.

[0083] The optimization unit is used to continuously evaluate and validate the model performance using the validation dataset during the training of large AI models, and to adjust and optimize the model parameters based on the evaluation results; it also introduces a meta-learning strategy to adapt to new scenarios and data changes.

[0084] Specifically, the AI ​​large-scale model training and optimization module first selects a suitable pre-trained model based on a deep learning framework and optimizes it for the urban rail transit field, constructing an AI large-scale model architecture adapted to the emergency management needs of urban rail transit depots. Then, the fused data is input into the model as training samples. Next, by designing reasonable loss functions (such as cross-entropy loss function, mean squared error loss function, etc.) and optimization algorithms (such as Adam optimization algorithm, gradient descent optimization algorithm, etc.), the model is trained iteratively in multiple rounds. During the training process, the dynamic adjustment of the learning rate is emphasized to ensure that the model can efficiently learn the data characteristics, correlation patterns, and emergency response rules of the depot under different operating states, such as normal and emergency situations.

[0085] Therefore, this invention introduces the following formula to dynamically adjust the learning rate, introducing hyperparameters. And construct the learning rate dynamic adjustment coefficient based on the loss function values ​​before and after training. The formula for dynamically adjusting the learning rate is:

[0086] (2);

[0087] (3);

[0088] in, This represents the current learning rate. This indicates the learning efficiency of the previous training session. Indicates the current training loss. This indicates the loss from the previous training session. This represents the dynamic adjustment coefficient for the current training. This represents the dynamic adjustment coefficient for the current training iteration. It is used for control The hyperparameters for adjusting the learning rate are set between 0 and 1. The above formula dynamically adjusts the learning rate based on changes in the model's loss, enhancing the flexibility and adaptability of the learning rate adjustment.

[0089] In this approach, the model can rapidly explore the parameter space with a relatively fast learning rate in the early stages of training. As training progresses, the learning rate is dynamically reduced based on changes in loss, allowing for fine-tuning as it approaches the optimal solution and ensuring stable convergence. Simultaneously, during model training, validation datasets are continuously used to evaluate and validate model performance, and model parameters are adjusted and optimized based on the evaluation results. Furthermore, the introduction of a meta-learning strategy enables the model to quickly adapt to new scenarios and data changes without requiring retraining the entire model. Ultimately, this ensures that the model achieves the expected performance indicators in terms of accuracy in emergency event identification, rationality of decision recommendations, and efficiency in resource allocation, meeting the practical application requirements of emergency management in urban rail transit depots.

[0090] The flowchart for the AI ​​large model training and optimization module is as follows: Figure 3 As shown.

[0091] In this embodiment, the emergency event intelligent identification and early warning module identifies emergency events based on a pre-trained AI model; and assesses the comprehensive risk value of the emergency events based on the identification results, issuing early warnings for emergency situations exceeding the risk threshold, including:

[0092] Step A: Based on the trained AI model, identify the emergency events and obtain the identification results; the identification results include the type, severity, scope of impact, and development trend of the emergency event.

[0093] Step B: Based on the identification results, the comprehensive risk value of the emergency event is assessed using the comprehensive risk calculation formula to obtain the comprehensive risk value of the emergency event;

[0094] The formula for calculating comprehensive risk is:

[0095] (4);

[0096] In the formula, This represents the overall risk value of an emergency event; Indicates the type of emergency event The severity level, ranging from 1 to 5; Indicates the type of emergency event The range of influence, with values ​​ranging from 1 to 5; Indicates the type of emergency event The probability of occurrence ranges from 0 to 1; This indicates the number of identified emergency event types;

[0097] Step C involves comparing the overall risk value of the emergency event with a preset risk threshold. If the overall risk value of the emergency event exceeds the preset risk threshold, an early warning signal is issued.

[0098] In the above technical solution, the emergency event intelligent identification and early warning module acquires data from various sensing devices and information systems at the depot in real time. First, it rapidly preprocesses and extracts features from the collected data, removing redundancy and noise interference to accurately extract key feature information related to the emergency event (such as train fault codes, abnormal environmental parameters, sudden changes in personnel density, etc.). Second, the extracted key feature data is input into a pre-trained AI model. Leveraging the depot's emergency knowledge and patterns learned by the model, it quickly and accurately identifies and judges the type (such as train malfunction, fire, flood, personnel fall, etc.), severity (such as mild, moderate, severe), impact range (such as local area, entire depot, surrounding traffic, etc.), and development trend (such as gradually worsening, remaining stable, gradually alleviating) of the current emergency event. Based on preset early warning rules, when the overall risk exceeds a preset threshold, an early warning signal is issued promptly to remind relevant personnel to take emergency measures.

[0099] The flowchart of the emergency event intelligent identification and early warning module is as follows: Figure 4 As shown.

[0100] In this embodiment, the intelligent response plan generation module, based on the comprehensive risk value determined by the intelligent identification results of the emergency event, provides emergency decision-makers with multiple targeted and operable intelligent response plan options from a massive database of historical emergency cases and expert experience knowledge, based on matching analysis of the characteristics of the current emergency event. Unlike the traditional model of using fixed response plans for emergency events, the intelligent response plan generation module of this invention can dynamically generate and optimize response plans according to the actual severity of the emergency situation at the scene, avoiding resource waste and significantly improving response efficiency. For example, in a fire scenario, response plans such as fire fighting and rescue, personnel evacuation, and activation of fire-fighting facilities are provided based on the size of the fire and the distribution of personnel. Simultaneously, a detailed assessment and analysis of the expected effects (such as recovery time, scope of impact, cost consumption, etc.), resource requirements (such as manpower, materials, equipment, etc.), and risk factors (such as secondary disaster risk, public opinion risk, etc.) of each response plan are conducted, assisting decision-makers in quickly selecting the optimal response plan based on the actual situation, thus achieving intelligent, scientific, and efficient emergency decision-making processes.

[0101] Specifically, the expression for the comprehensive evaluation of the disposal plan is as follows:

[0102] (5);

[0103] in, This indicates the overall evaluation score of the disposal plan. This represents the score of the treatment plan on the effectiveness dimension k (value range 0 to 100). Indicates the weight of the effect dimension k (value range from 0 to 1); This represents the score of the solution on the risk dimension l (with a value range of 0 to 100). This represents the weight of risk dimension l (ranging from 0 to 1). ; Indicates the number of effect dimensions. This indicates the number of risk dimensions.

[0104] The flowchart of the intelligent generation module for handling solutions is as follows: Figure 5 As shown.

[0105] In this embodiment, the emergency process assessment and model optimization module collects and organizes various data and information during the emergency process, including emergency response time, resource allocation efficiency, implementation effect of the response plan, material consumption, and equipment operating status, to construct a multi-dimensional emergency response effectiveness evaluation index system. Using scientific data analysis methods and evaluation models, the overall effectiveness of this emergency response work is quantitatively evaluated. The evaluation results serve as an important reference for the optimization and training of the AI ​​large-scale model. Combined with newly collected emergency case data, the AI ​​large-scale model undergoes continuous incremental learning and optimization training, constantly improving its understanding and perception of various emergency scenarios in urban rail transit sections. This enhances the model's performance in emergency scenario identification, decision-making suggestions, and resource allocation, achieving continuous improvement and optimization of the emergency management system's performance to better cope with various future emergency events.

[0106] Specifically, the expression for evaluating the effectiveness of emergency response is:

[0107] (6);

[0108] in, This indicates the score for evaluating the effectiveness of the emergency response; Indicates the emergency response time; This indicates the maximum emergency response time according to the emergency management regulations for the site. Indicates resource utilization efficiency; This indicates the maximum resource utilization rate according to the emergency management regulations for the site. This indicates the score for evaluating the effectiveness of the emergency response; This indicates the maximum emergency response score based on the emergency management regulations for the site. This indicates the total cost of emergency resources; This indicates the maximum total cost of emergency resources according to the emergency management regulations for the site. , , , These represent the weighting coefficients for emergency response time, resource utilization rate, emergency effectiveness evaluation score, and total emergency resource cost, respectively. .

[0109] The flowchart for the emergency process assessment and model optimization module is as follows: Figure 6 As shown.

[0110] In practice, the following example illustrates the emergency response to a fire at a rail transit depot in a Chinese city:

[0111] (1) Multi-source data acquisition and fusion processing module

[0112] This involves monitoring key data related to fire incidents at the site, including smoke concentration detected by smoke sensors. ), the temperature detected by the temperature sensor ( ), the area affected by the fire as captured by video surveillance equipment ( );

[0113] The comprehensive fire risk index, reflecting smoke concentration, temperature, and fire-affected area, is calculated using the above formula (1). Among these factors, , , Corresponding to , , ; , Based on the regulations governing fire emergency response at the site, the normal smoke concentration and the upper limit of smoke concentration at the site are specified. , , These represent the risk assessment values ​​for smoke concentration, temperature, and fire impact area, respectively, as assessed according to the site's emergency management regulations. This represents the maximum fire risk assessment value that the site can withstand. ; , , Risk amplification coefficients for different dimensions of fire parameters can be calculated based on historical fire event statistics. This is the time decay coefficient for fire emergency events, which can be determined based on the event type of the current fire. , These are the duration of a fire in the site and the maximum duration of a fire that the site can withstand.

[0114] The multi-source data acquisition and fusion processing module comprehensively assesses the risk index of fire emergency events by considering fire-related factors such as smoke concentration, temperature, and the area affected by the fire, thus quantifying fire risk. This formula-based fusion method not only comprehensively reflects the urgency of fire events but also dynamically adjusts the risk assessment results to adapt to real-time changes in fire development. By integrating data from different dimensions into a comprehensive risk index, this module provides emergency management personnel with an intuitive and comprehensive risk assessment tool, helping to quickly identify the severity and development trend of fires, thereby enabling timely and effective emergency measures. Compared with traditional methods, this approach can more accurately assess fire risk, improve the efficiency and effectiveness of emergency response, and play a crucial role in ensuring the safe operation of urban rail transit depots. The fused data is stored in a structured depot emergency management data knowledge base, providing high-quality data support for subsequent AI large-scale model training.

[0115] (2) AI Large Model Training and Optimization Module

[0116] This study utilizes a deep learning framework and pre-trained models for adaptation and optimization in the urban rail transit field. It constructs a large-scale AI model architecture adapted to the emergency management needs of urban rail transit depots. Data processed by a multi-source data acquisition and fusion module is used as training samples, along with historical emergency event case data, expert experience data, and depot emergency event management data, as input to the model. During training, a reasonable loss function and optimization algorithm are designed to iterate the model through multiple rounds of training, with an initial learning rate set at a certain threshold. The current loss value during training is The previous loss value was The learning rate of the AI ​​large model is adjusted according to Equations (2) and (3) to optimize the training of the model.

[0117] (3) Emergency Event Intelligent Identification and Early Warning Module

[0118] Leveraging the emergency response knowledge and patterns learned by the AI ​​large-scale model, the current emergency event type (fire, etc.) can be analyzed. (indication), severity () ), Scope of impact ( ) and development probability ( ) to quickly and accurately identify and assess, and use formula (4) to calculate the current emergency event risk value ( According to the pre-set early warning rules in the site management regulations, when the risk value of an emergency event exceeds the preset threshold ( When an emergency occurs, the system will promptly issue a warning signal to remind relevant personnel to take emergency measures.

[0119] (4) Intelligent generation module for disposal plan

[0120] Utilizing an AI big data model, information on fire-related response plans is retrieved from a massive database of historical cases and expert experience. Based on the current event's characteristic data (smoke concentration, temperature, fire-affected area, severity, scope of impact, and probability of development), the AI ​​big data model intelligently generates multiple targeted and actionable intelligent response plan options. For the current fire incident, the AI ​​big data model recommends and generates several response plans. Each response plan is evaluated based on fire extinguishing efficiency (…). ), personnel evacuation ), and prevention of secondary disasters from fires ( The three performance dimensions are evaluated to obtain specific scores. , , and various weights , , The score in the effectiveness dimension can then be expressed as: Each response plan considers personnel safety risks ( Resource allocation risk ), environmental uncertainty risk ( The three risk dimensions are assessed to obtain specific scores. , , and various weights , , The score in the risk dimension can then be expressed as: Equation (5) is used to comprehensively evaluate the performance of each scheme in terms of both effectiveness and risk, resulting in a comprehensive evaluation score for each scheme. The effect dimension influences the weight ( ), )and( The optimal response plan can be determined based on the management priorities of the site for fire emergency incidents. By comparing the comprehensive evaluation values ​​of various plans, the best response plan can be finally determined.

[0121] (5) Emergency Process Assessment and Model Optimization Module

[0122] Collect and organize relevant data during the implementation of the emergency response plan, including emergency response time, resource utilization efficiency, effectiveness of the response plan, and material consumption. Assess the response time for emergency incidents according to the site's emergency management regulations. Resource utilization efficiency in emergency response ( ), emergency response effectiveness ( ), total cost of emergency resource consumption ( Using formula (6), the overall evaluation score of emergency response effectiveness is obtained by quantitatively assessing the emergency response effectiveness, which is jointly reflected by emergency response time, resource utilization efficiency, emergency response effect, and total cost of emergency resource consumption. The comprehensive evaluation values ​​reflect the effectiveness of this emergency response and the rationality of resource utilization, providing crucial data support and quantitative basis for the continuous optimization of the AI ​​big data model.

[0123] By comprehensively evaluating the effectiveness of emergency response, the strengths and weaknesses of the process can be accurately identified. This information is integrated into the model's feedback mechanism, driving incremental learning and parameter adjustments. Based on the evaluation results, the model strengthens its ability to identify efficient response plans and improves inefficient decision-making logic, thereby enhancing its intelligence in emergency scenario identification, decision suggestion generation, and effectiveness evaluation. With continuous optimization, the model can more accurately predict the optimal response strategy in similar scenarios, significantly improving the response speed, decision-making scientificity, and resource utilization efficiency of urban rail transit depot emergency management systems in the face of future emergencies.

[0124] Example 2

[0125] like Figure 7 As shown, the difference between this embodiment and Embodiment 1 is that this embodiment further provides an emergency management method for urban rail transit depots based on an AI large-scale model. This method corresponds one-to-one with the emergency management system for urban rail transit depots based on an AI large-scale model in Embodiment 1. The method includes:

[0126] Acquire multi-dimensional detection data related to emergency events within the field area, and then weight and fuse the multi-dimensional detection data with relevant factors to obtain the fused data;

[0127] The fused data, historical emergency event case data, expert experience data, and field emergency event management data are used as training samples and input into the pre-built AI big model. The AI ​​big model is trained by dynamically adjusting the learning rate to obtain a trained AI big model.

[0128] Based on real-time acquired multi-dimensional detection data, key feature information related to emergency events is extracted and input into a pre-trained AI model. The pre-trained AI model is then used to identify emergency events. The comprehensive risk value of emergency events is assessed based on the identification results, and warnings are issued for emergency situations that exceed the risk threshold.

[0129] Based on the comprehensive risk value of the emergency, multiple response plans are generated using a pre-trained AI model, and a comprehensive evaluation of these plans is conducted to generate the optimal response plan.

[0130] As a further implementation, the method also includes:

[0131] Based on relevant data during the emergency response, an evaluation of the emergency response effectiveness is conducted; and the evaluation results and emergency response process data are fed back to the AI ​​big model for model optimization and updates.

[0132] The implementation of each step can be carried out in accordance with the execution process of each unit of the urban rail transit depot emergency management system based on the AI ​​large model in Example 1, and will not be described in detail in this example.

[0133] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0134] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0135] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0136] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1The steps of the function specified in one or more boxes.

[0137] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. An urban rail transit section emergency management system based on an AI large model, characterized in that, The system comprises: A multi-source data acquisition and fusion processing module for acquiring multi-dimensional detection data related to emergency events in a field section, and weighting and fusing the multi-dimensional detection data with related factors to obtain fused data; The expression of the weighted fusion is: ； wherein, is the fused data; denotes the emergency characteristic parameter at time is the real-time value of the emergency characteristic parameter , are respectively the reference value and the upper threshold value of the emergency characteristic parameter ; is the risk assessment value of the emergency characteristic parameter ; is the maximum value of ; , , are respectively the weight coefficient, the risk enhancement coefficient and the time decay coefficient; and are respectively the duration of the emergency event and the maximum duration that the scene section can withstand for the emergency event; is the data dimension involved in the emergency event; An AI large model training and optimization module for inputting the fused data, historical emergency event case data, expert experience data and field section emergency event management data into a pre-constructed AI large model as training samples, training the AI large model by combining a dynamically adjusted learning rate, and obtaining a trained AI large model; An emergency event intelligent identification and early warning module for extracting key feature information related to emergency events from real-time acquired multi-dimensional detection data, inputting the key feature information into the trained AI large model, identifying the emergency events based on the trained AI large model, and evaluating the comprehensive risk value of the emergency events based on the identification results to perform early warning on emergency situations exceeding a risk threshold; Based on the trained AI large model, the emergency events are identified, and the comprehensive risk value of the emergency events is evaluated based on the identification results to perform early warning on emergency situations exceeding a risk threshold, including: Based on the trained AI large model, the emergency events are identified, and the comprehensive risk value of the emergency events is evaluated based on the identification results to perform early warning on emergency situations exceeding a risk threshold, including: In combination with the recognition result, a comprehensive risk calculation formula is used to evaluate the comprehensive risk value of the emergency event, so as to obtain the comprehensive risk value of the emergency event; the comprehensive risk calculation formula is: , wherein, represents the comprehensive risk value of the emergency event; represents the severity of the emergency event type , and the value range is 1 to 5; represents the influence range of the emergency event type , and the value range is 1 to 5; represents the occurrence probability of the emergency event type , and the value range is 0 to 1; represents the number of the recognized emergency event types. The comprehensive risk value of the emergency events is compared with a preset risk threshold, and when the comprehensive risk value of the emergency events exceeds the preset risk threshold, an early warning signal is sent out; A disposal scheme intelligent generation module for generating multiple disposal schemes based on the trained AI large model according to the comprehensive risk value of the emergency events, and performing disposal scheme comprehensive evaluation to generate an optimal disposal scheme; The expression of the disposal scheme comprehensive evaluation is: ； wherein, represents a composite evaluation score of the treatment regimen, represents a score of the treatment regimen on the effect dimension k; represents a weight of the effect dimension k; represents a score of the treatment regimen on the risk dimension l, represents a weight of the risk dimension l, ; represents a number of effect dimensions, represents a number of risk dimensions.

2. The AI large model-based urban rail transit section emergency management system according to claim 1, characterized in that, The system further comprises: An emergency process evaluation and model optimization module for performing emergency disposal effect evaluation according to related data in the emergency process, and feeding back the evaluation results and emergency disposal process data to the AI large model for model optimization and updating. 3.The AI large model-based urban rail transit section emergency management system according to claim 1, wherein, The multi-source data acquisition and fusion processing module comprises: A multi-source data acquisition unit for classifying emergency events according to pre-processed emergency event types in a field section, and acquiring multi-dimensional detection data related to emergency events in the field section for each type of emergency event; the multi-dimensional detection data includes temperature data, smoke concentration data and video data of fire accidents; A fusion processing unit for pre-processing the multi-dimensional detection data to obtain pre-processed feature parameters, weighting and fusing the pre-processed feature parameters with related factors, and incorporating a time factor in the fusion process to obtain fused data. 4.The AI large model-based urban rail transit section emergency management system according to claim 1, wherein, The AI large model training and optimization module comprises: An AI large model training unit for inputting the fused data, historical emergency event case data, expert experience data and field section emergency event management data into a pre-constructed AI large model as training samples, training the AI large model by combining a dynamically adjusted learning rate, and obtaining a trained AI large model; An optimization unit is configured to continuously evaluate and verify the performance of the model using the validation dataset during the AI large model training process, and adjust and optimize the model parameters according to the evaluation results; a meta-learning strategy is also introduced to adapt to new scenarios and data changes.

5. The AI large model-based urban rail transit section emergency management system according to claim 1 or 4, characterized in that, The dynamic learning rate is introduced in the AI large model training process , and a learning rate dynamic adjustment coefficient is constructed according to the loss function values before and after training ; the expression of the dynamic learning rate is: ； ； wherein, denotes the current learning rate, denotes the learning efficiency of the previous training, denotes the current training loss, denotes the previous training loss, denotes the dynamic adjustment coefficient of the current training, denotes the dynamic adjustment coefficient of the current training, is an hyperparameter for controlling the adjustment speed, and the value range is between 0 and 1. 6.The AI large model-based urban rail transit section emergency management system according to claim 2, wherein, The expression of the emergency treatment effect evaluation is: ； wherein, represents an emergency handling effect evaluation score; represents an emergency response time; represents a maximum emergency response time according to the field section emergency management regulation; represents a resource utilization efficiency; represents a maximum resource utilization rate according to the field section emergency management regulation; represents an emergency effect evaluation score; represents a maximum emergency effect evaluation score according to the field section emergency management regulation; represents an emergency resource total cost; represents a maximum emergency resource total cost according to the field section emergency management regulation; , , , respectively represent a weight coefficient of the emergency response time, the resource utilization rate, the emergency effect evaluation score, and the emergency resource total cost, .

7. The urban rail transit section emergency management method based on an AI large model, characterized in that, The method comprises: Obtain multi-dimensional detection data related to emergency events in the field section, and weight and fuse the multi-dimensional detection data and related factors to obtain fused data; the expression of the weighted fusion is: ； wherein, is the fused data; denotes the emergency characteristic parameter at the time point ; , are respectively the reference value and the upper threshold value of the emergency characteristic parameter ; is the risk assessment value of the emergency characteristic parameter ; is the maximum value of ; , , are respectively the weight coefficient, the risk enhancement coefficient and the time decay coefficient; and are respectively the duration of the emergency event and the maximum duration that the scene section can withstand for the emergency event; is the data dimension involved in the emergency event; The fused data, historical emergency event case data, expert experience data and field section emergency event management data are input into the pre-built AI large model as training samples, and the AI large model is trained by combining a dynamically adjusted learning rate, to obtain a trained AI large model; According to the real-time obtained multi-dimensional detection data, key feature information related to the emergency event is extracted, and the key feature information is input into the trained AI large model, and the trained AI large model is used to identify the emergency event; and the comprehensive risk value of the emergency event is evaluated according to the identification result, and the emergency situation exceeding the risk threshold is warned; Based on the trained AI large model, the emergency event is identified; and the comprehensive risk value of the emergency event is evaluated according to the identification result, and the emergency situation exceeding the risk threshold is warned, including: Based on the trained AI large model, the emergency event is identified, and the identification result is obtained; the identification result includes the type, severity, influence range and development trend of the emergency event; In combination with the recognition result, a comprehensive risk calculation formula is used to evaluate the comprehensive risk value of the emergency event, so as to obtain the comprehensive risk value of the emergency event; the comprehensive risk calculation formula is: , wherein, represents the comprehensive risk value of the emergency event; represents the severity of the emergency event type , and the value range is 1 to 5; represents the influence range of the emergency event type , and the value range is 1 to 5; represents the occurrence probability of the emergency event type , and the value range is 0 to 1; represents the number of the recognized emergency event types. The comprehensive risk value of the emergency event is compared with the preset risk threshold, and when the comprehensive risk value of the emergency event exceeds the preset risk threshold, a warning signal is sent; According to the comprehensive risk value of the emergency event, a plurality of treatment schemes are generated based on the trained AI large model, and treatment scheme comprehensive evaluation is performed to generate an optimal treatment scheme; The expression of the treatment scheme comprehensive evaluation is: ； wherein, represents a composite evaluation score of the treatment regimen, represents a score of the treatment regimen on the effect dimension k; represents a weight of the effect dimension k; represents a score of the treatment regimen on the risk dimension l, represents a weight of the risk dimension l, ; represents a number of effect dimensions, represents a number of risk dimensions.

Images