A method and system for adaptively generating emergency strategies for sudden incidents in highway tunnels

Abstract

The present invention discloses a method and system for adaptively generating emergency strategies for sudden incidents in highway tunnels, which relates to the field of intelligent traffic control technology. The key points of the technical solution are: collecting multimodal data; extracting event basic features, tunnel structural features, and environmental dynamic features layer by layer through a dynamic feature hierarchical extraction module to generate a multidimensional feature vector; performing event type primary screening, severity secondary screening, and environmental adaptation verification in sequence through a control item dynamic screening module to screen candidate control items from a total control item set; performing multi-objective optimization on the candidate control items to generate a Pareto optimal strategy set, and outputting the final emergency strategy in conjunction with an interactive decision support interface. The present invention can effectively improve the accuracy of event type identification and the timeliness of environmental risk prediction. Through the three-stage verification of the control item dynamic screening module, the candidate control item screening accuracy is increased, effectively avoiding strategy conflicts and resource waste.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent traffic control, and more particularly to a method and system for adaptively generating emergency strategies for sudden incidents in highway tunnels. Background Art

[0002] As a vital component of transportation infrastructure, highway tunnels present a high risk of emergencies such as fires and traffic accidents due to their enclosed spatial structures and complex operating environments. These emergencies often exhibit characteristics such as multi-factor coupling, rapid dynamic evolution, and high risk of secondary disasters. Traditional emergency decision-making relies on manual experience and static emergency plan libraries, resulting in delayed response, poor environmental adaptability, and difficulty coordinating conflicting objectives.

[0003] With the widespread adoption of IoT sensing technology and intelligent algorithms, building a data-driven, adaptive emergency strategy generation system has become a key approach for improving tunnel safety operations. Prior art describes emergency decision-making frameworks based on fuzzy logic, but these utilize a single sensor data fusion approach with a single feature extraction dimension, making them ineffective in processing the spatiotemporal correlations of multimodal data, such as capturing the synergistic effects of key parameters like smoke diffusion gradients and visibility attenuation rates. Some prior art also utilizes genetic algorithms for strategy optimization, but lacks a feature stratification mechanism, resulting in inadequate modeling of the coupling relationship between dynamic environmental parameters and structural constraints. Furthermore, these prior art emergency strategy generation methods primarily target emergencies on highways. When applied to highway tunnels, the control item screening process lacks a verification mechanism, which can easily lead to mismatches between ventilation strategies and fluid dynamics characteristics, and conflicts between evacuation routes and crowd dynamics models.

[0004] Therefore, how to research and design a method and system for adaptively generating emergency strategies for highway tunnel emergencies that can overcome the above-mentioned defects is a problem that we urgently need to solve. Summary of the Invention

[0005] In order to address the deficiencies in the prior art, the purpose of the present invention is to provide a method and system for adaptively generating emergency strategies for sudden incidents in highway tunnels. A multidimensional feature vector is constructed through a dynamic feature hierarchical extraction module, and a three-level feature fusion mechanism is combined with the event basic feature layer, tunnel structure feature layer and environmental dynamic feature layer. Compared with the traditional single-layer feature extraction method, the accuracy of event type recognition and the timeliness of environmental risk prediction can be effectively improved. Through the three-stage verification of the control item dynamic screening module, the screening accuracy of candidate control items is made higher, effectively avoiding strategy conflicts and resource waste.

[0006] The above technical objectives of the present invention are achieved through the following technical solutions:

[0007] In a first aspect, a method for adaptively generating an emergency response strategy for a highway tunnel incident is provided, comprising the following steps:

[0008] Collect sensor data, video surveillance data, and environmental parameters in the target tunnel to obtain multimodal data;

[0009] Based on the multimodal data, a dynamic feature hierarchical extraction module is used to extract event basic features, tunnel structure features, and environmental dynamic features layer by layer to generate a multidimensional feature vector;

[0010] Based on the multidimensional feature vector, the control item dynamic screening module sequentially performs event type primary screening, severity secondary screening, and environmental adaptation verification to screen out candidate control items from the total control item set;

[0011] The NSGA-III algorithm or the improved NSGA-III algorithm is used to perform multi-objective optimization on the candidate control items to generate a Pareto optimal strategy set, and the final emergency strategy is output in combination with an interactive decision support interface.

[0012] Furthermore, the dynamic feature hierarchical extraction module includes:

[0013] The event basic feature layer is used to extract event basic features including temperature, smoke concentration, CO concentration, vehicle density and accident point location data;

[0014] Tunnel structural feature layer, used to extract tunnel structural features including tunnel curvature, ventilation shaft distribution, escape route topology, and fire protection facility layout;

[0015] The environmental dynamic feature layer is used to extract environmental dynamic features including wind speed vector field, temperature and humidity gradient matrix and visibility attenuation rate.

[0016] Furthermore, the event type initial screening is achieved using a multimodal fusion classification model;

[0017] Among them, the modal fusion classification model takes the event basic features and the tunnel structure features as input, performs feature splicing operation on the event basic features and the tunnel structure features, and applies a multi-classification activation function to output the probability distribution vector of the event type.

[0018] Furthermore, the secondary screening of severity is achieved by calculating a risk index, specifically including:

[0019] Solving the environmental dynamic characteristics inversely in time to obtain the environmental change risk;

[0020] Normalizing the basic features of the event and calculating the L2 norm to obtain the global event risk;

[0021] The risk index is determined by taking the environmental change risk and the sum of the environmental change risk.

[0022] Furthermore, the environmental adaptation verification includes:

[0023] Perform fluid dynamics simulation verification on ventilation system controls;

[0024] Perform crowd dynamics model validation on evacuation path planning items;

[0025] And / or, perform reachability topology analysis on fire rescue deployment items.

[0026] Furthermore, the objective function of the multi-objective optimization uses the activation state of the candidate control item as a decision variable;

[0027] The objective function includes:

[0028] Minimize the time objective, including the coupled effects of control item execution time and dynamic environment gradients;

[0029] Minimize cost objectives and introduce opportunity costs to quantify the losses caused by abandoning other measures due to enabling control items;

[0030] Minimize the risk objective and use the product form to represent the complex risk amplification effect brought about by the superposition of multiple control items;

[0031] Constraints include mutually exclusive control item constraints, dependent control item constraints, and tunnel physical constraints.

[0032] Furthermore, the improvements of the improved NSGA-III algorithm include:

[0033] Guiding the search direction through a dynamic reference point and adaptively dividing the hyperplane according to the dimensions of the candidate control items;

[0034] Furthermore, a constraint violation weighting strategy is adopted to penalize solutions that do not satisfy the tunnel structure constraints.

[0035] Furthermore, the interactive decision support interface outputs a final emergency strategy, including:

[0036] The minimum Manhattan distance solution is calculated in the Pareto optimal strategy set as the recommended strategy, and the final emergency strategy is output.

[0037] Furthermore, the method further comprises:

[0038] Dynamically updating a feature extraction rule base and weight parameters of the multi-objective optimization according to the execution results of the final emergency strategy;

[0039] The dynamically updated feature extraction rule base adopts an incremental reinforcement learning mechanism.

[0040] In a second aspect, a system for adaptively generating emergency strategies for sudden incidents in highway tunnels is provided. The system is configured to implement the method for adaptively generating emergency strategies for sudden incidents in highway tunnels as described in any one of the first aspects, comprising:

[0041] The data acquisition module is used to collect sensor data, video surveillance data and environmental parameters in the target tunnel to obtain multimodal data;

[0042] A feature extraction module is used to extract event basic features, tunnel structure features and environmental dynamic features layer by layer based on the multimodal data through a dynamic feature hierarchical extraction module to generate a multidimensional feature vector;

[0043] A control screening module is used to perform event type primary screening, severity secondary screening and environment adaptation verification in sequence through a control item dynamic screening module based on the multi-dimensional feature vector, and screen out candidate control items from the total control item set;

[0044] The target optimization module is used to perform multi-objective optimization on the candidate control items using the NSGA-III algorithm or the improved NSGA-III algorithm, generate a Pareto optimal strategy set, and output the final emergency strategy in combination with an interactive decision support interface.

[0045] Compared with the prior art, the present invention has the following beneficial effects:

[0046] 1. The present invention provides a method for adaptively generating emergency response strategies for highway tunnel emergencies. This method constructs a multidimensional feature vector through a dynamic feature layered extraction module. This method combines a three-level feature fusion mechanism consisting of an event basic feature layer, a tunnel structure feature layer, and an environmental dynamic feature layer. Compared to traditional single-layer feature extraction methods, this method can effectively improve the accuracy of event type identification and the timeliness of environmental risk prediction. Furthermore, through a three-stage verification process using a dynamic control item screening module, the method achieves higher precision in screening candidate control items, effectively avoiding policy conflicts and resource waste.

[0047] 2. This invention uses an improved NSGA-III algorithm to achieve multi-objective optimization. By guiding the search direction and constraint violation weighting strategy through dynamic reference points, it can accelerate the convergence of the Pareto front in multi-dimensional control item scenarios and improve the coverage of the strategy set. In addition, combined with the minimum Manhattan distance solution recommendation mechanism, the decision-making efficiency is higher than that of the traditional Euclidean distance method.

[0048] 3. The present invention introduces fluid mechanics simulation and crowd dynamics model for joint verification in the environmental adaptation verification link, which can improve the smoke dispersal efficiency of the ventilation strategy and shorten the detention time of personnel in evacuation path planning; and optimizes the fire rescue path through accessibility topology analysis, which can reduce the response time of emergency resource deployment. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] The drawings described herein are used to provide a further understanding of the embodiments of the present invention, constitute a part of this application, and do not constitute a limitation of the embodiments of the present invention. In the drawings:

[0050] Figure 1 This is a flowchart of Example 1 of the present invention;

[0051] Figure 2 This is a system block diagram in Example 2 of the present invention. DETAILED DESCRIPTION

[0052] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with examples and drawings. The exemplary embodiments of the present invention and their descriptions are only used to explain the present invention and are not intended to limit the present invention.

[0053] Example 1: A method for adaptively generating emergency response strategies for highway tunnel emergencies, such as Figure 1 As shown, the following steps are included:

[0054] S1: Collect sensor data, video surveillance data, and environmental parameters in the target tunnel to obtain multimodal data;

[0055] S2: Based on multimodal data, the dynamic feature hierarchical extraction module extracts event basic features, tunnel structure features, and environmental dynamic features layer by layer to generate a multidimensional feature vector.

[0056] S3: Based on the multi-dimensional feature vector, the control item dynamic screening module performs event type primary screening, severity secondary screening, and environmental adaptation verification in sequence to select candidate control items from the total control item set;

[0057] S4: Use the NSGA-III algorithm or the improved NSGA-III algorithm to perform multi-objective optimization on the candidate control items, generate a Pareto optimal strategy set, and output the final emergency strategy through an interactive decision support interface. The NSGA-III algorithm is a third-generation non-dominated sorting genetic algorithm.

[0058] In step S1, sensor data is the core data for real-time monitoring of the tunnel environment status, which is collected through distributed IoT devices, including but not limited to temperature, smoke concentration, carbon monoxide concentration, carbon dioxide concentration, oxygen concentration, harmful gases, vibration frequency of the tunnel lining, crack width, water seepage pressure, vehicle density, vehicle speed and accident point location information.

[0059] Video surveillance data is a key input for visual perception and event recognition. It extracts information through multi-source cameras and intelligent analysis algorithms, including but not limited to vehicle accident types, number of stranded personnel, visible light recognition of flames / smoke, thermal radiation distribution in high-temperature areas, detection of hidden fire sources, three-dimensional outlines of obstacles, spatial location of vehicles / personnel, detection of running / falling personnel, and judgment of vehicles driving in the wrong direction / illegally parked.

[0060] Environmental parameters are auxiliary data for assessing dynamic risks and strategy feasibility. They are combined with real-time meteorological and equipment status collection, including but not limited to wind speed / direction, humidity, air pressure, light intensity, fan speed, fire water pressure, remaining power of emergency power supply, wireless signal strength and other information.

[0061] In step S2, the dynamic feature hierarchical extraction module includes an event-based feature layer, a tunnel structure feature layer, and an environmental dynamic feature layer. This invention utilizes a three-level hierarchical extraction mechanism to address the heterogeneity of multimodal data and the dynamic risk evolution characteristics of highway tunnel emergency scenarios. The event-based feature layer primarily extracts features directly reflecting the nature of the event from the raw data; the tunnel structure feature layer primarily loads static tunnel topology information to ensure policy compliance with physical constraints; and the environmental dynamic feature layer predicts environmental evolution trends to achieve forward-looking policy optimization.

[0062] Specifically, the event basic feature layer is used to extract basic event features such as temperature, smoke concentration, CO concentration, vehicle density and accident point location data; the tunnel structure feature layer is used to extract tunnel structure features such as tunnel curvature, ventilation shaft distribution, escape channel topology and fire protection facility layout; the environmental dynamic feature layer is used to extract environmental dynamic features such as wind speed vector field, temperature and humidity gradient matrix and visibility attenuation rate.

[0063] For example, the data input of the event basic feature layer is: temperature (95℃), smoke concentration (800μg / m³), CO concentration (1200ppm), vehicle density (12 vehicles / 100m), calculate the mean / extreme value (such as maximum temperature, maximum smoke concentration) within 50m around the accident point, fuse multi-sensor data through Kalman filtering, eliminate noise and generate a continuous spatiotemporal distribution map, and obtain the event basic features. .

[0064] The data input of the tunnel structure characteristic layer is: BIM model (the accident point is located at the K23+500 bend section), ventilation shaft location (upstream well distance 350m, downstream well distance 200m), the number of available fire hydrants (2, water pressure 0.8MPa), calculation of the shortest path between the accident point and key facilities (such as the distance to the escape route), definition of the maximum allowable fire truck width in the bend area (3.5m), and the resulting tunnel structure characteristics. .

[0065] The data input of the environmental dynamic feature layer is: wind speed vector (2.8m / s, direction S→N), temperature and humidity gradient (temperature rise rate 8℃ / min, humidity drop rate 15% / min), visibility attenuation rate (-15% / min), based on fluid dynamics simulation to predict the smoke diffusion path, using long short-term memory network to estimate the environmental state in the next 5 minutes, the obtained environmental dynamic features .

[0066] After obtaining the event basic features, tunnel structure features and environmental dynamic features in layers, the three layers of features are integrated into a unified vector through feature level fusion, which is the multidimensional feature vector .

[0067] In highway tunnel fire scenarios, dynamic feature layered extraction transforms raw data into feature vectors with clear physical meaning through multi-source data fusion and layered processing. Event-based features locate core accident parameters and determine response levels; tunnel structural features constrain strategy feasibility and optimize resource deployment; and environmental dynamic features predict event trends and enable proactive control. This method enables rapid feature extraction in complex environments for emergency strategy generation, providing high signal-to-noise ratio input for subsequent multi-objective optimization, significantly improving the efficiency and reliability of emergency response.

[0068] In step S3, the initial event type screening filters out irrelevant control items based on the event type (e.g., fire, traffic accident). For example, a fire might require smoke exhaust and lane closures, while a traffic accident might require route diversion and lighting control. The event type is determined by combining basic event features from the multidimensional feature vector, such as temperature and smoke density.

[0069] In some examples, the initial screening of event types is achieved using a multimodal fusion classification model; wherein the modal fusion classification model takes the basic features of the event and the tunnel structure features as input, performs feature splicing operations on the basic features of the event and the tunnel structure features, and then applies a multi-classification activation function to output the probability distribution vector of the event type.

[0070] Specifically, the expression of the modal fusion classification model is:

[0071] ;

[0072] in, A probability distribution vector representing the event type, such as the predicted probability of fire, traffic accident, or hazardous material leakage; Represents a multi-classification activation function; Represents a trainable weight matrix whose dimension is the product of the number of event types and the total dimension of features; Indicates the basic characteristics of the event; Indicates the structural characteristics of the tunnel; represents the bias term; Represents a feature concatenation operation.

[0073] Secondary severity screening further narrows the scope of control measures based on the severity of the incident. For example, a minor fire might require only partial lane closures, while a severe fire might require full closures. Severity assessment may require a risk index formula that incorporates dynamic characteristics such as smoke concentration gradients and temperature change rates.

[0074] In some examples, secondary screening of severity is achieved through risk index calculation, specifically including: solving the time inverse of the dynamic characteristics of the environment to obtain the environmental change risk; normalizing the basic characteristics of the event and calculating the L2 norm to obtain the global event risk; and determining the risk index as the sum of the environmental change risk and the environmental change risk.

[0075] The specific expression of the risk index is:

[0076] ;

[0077] in, represents the risk index; Represents the environmental dynamic sensitivity coefficient, which is used to amplify the impact of environmental mutations; This index represents the basic sensitivity coefficient of an event, used to measure the severity of the event itself. This invention triggers a significant response when the risk index exceeds a set threshold. For example, in the event of a fire, key actions such as full power operation of the smoke exhaust system and two-way traffic control can be activated.

[0078] Environmental suitability verification ensures that the selected control items are compatible with the current environmental dynamics and tunnel structure. For example, CFD simulation can be used to verify the feasibility of ventilation strategies or to check whether evacuation routes are unavailable due to structural issues. This verification requires combining tunnel structural characteristics (such as ventilation shaft location) with environmental dynamic characteristics (such as wind speed and visibility).

[0079] In some examples, environmental adaptation verification can be performed in one or more of the following ways: performing fluid dynamics simulation verification on ventilation system control items; performing crowd dynamics model verification on evacuation path planning items; and performing reachability topology analysis on fire rescue deployment items.

[0080] In some examples, performing fluid dynamics simulation verification on ventilation system control items is mainly to verify whether ventilation strategies such as smoke exhaust and air supply comply with the laws of smoke diffusion, to prevent strategy failure or increase risks.

[0081] A 3D tunnel model was built using computational fluid dynamics (CFD) and loaded with real-time parameters: structural and dynamic parameters. Structural parameters included, but were not limited to, tunnel cross-sectional dimensions, ventilation shaft location, and fan power; dynamic parameters included, but were not limited to, real-time wind speed and smoke concentration gradient. By setting boundary conditions and simulating smoke diffusion under different ventilation strategies, key indicators such as visibility recovery time and the rate of reduction of high-temperature areas were determined.

[0082] The boundary condition expression is:

[0083] ;

[0084] in, represents the velocity field; Indicates time; represents the convection acceleration; Indicates the air density; represents the pressure gradient; Indicates the kinematic viscosity of air; represents the Laplace of the velocity field; Indicates smoke buoyancy.

[0085] In some examples, crowd dynamics model verification is performed on evacuation path planning items to ensure that the planned escape path meets the crowd behavior characteristics and dynamic environment constraints to prevent trampling or congestion.

[0086] For example, the path planning solution includes guiding to horizontal tunnel H3 (3.2m wide); real-time data includes: crowd density (2 people / m²) and visibility (15m); dynamic simulation predicts the travel time for people through horizontal tunnel H3 within 5 minutes and detects bottleneck areas, such as curves where the speed drops by 30% due to limited vision.

[0087] In some examples, performing reachability topology analysis on fire rescue deployment items is mainly to verify the physical reachability and timeliness of fire resource dispatch paths to ensure the rapid arrival of rescue forces.

[0088] For example, a tunnel space topology map is constructed based on the BIM model. The nodes in the tunnel space topology map are: accident point (K23+500), fire hydrant (HYD1 / 2), and cross tunnel (H3 / H4); edge weight is: travel distance × dynamic risk coefficient, such as the weight +50% when the CO concentration is >1000ppm; then the obstacle is updated in real time: the landslide area (coordinates K23+480→K23+520); the optimal path is solved through the algorithm: the fire truck detours from the entrance → cross tunnel H4.

[0089] In step S4, the objective function of the multi-objective optimization uses the activation state of the candidate control item as the decision variable.

[0090] The objective functions include: minimizing the time target, which includes the coupled impact of the execution time of the control item and the dynamic environmental gradient; minimizing the cost target, which introduces the opportunity cost and quantifies the loss caused by abandoning other measures due to the activation of the control item; minimizing the risk target, which uses the product form to characterize the complexity risk amplification effect brought about by the superposition of multiple control items.

[0091] Constraints include mutually exclusive control item constraints, dependent control item constraints, and tunnel physical constraints.

[0092] Specifically, the expression of the objective function is:

[0093] ;

[0094] in, represents the minimization time objective; represents a set of binary decision variables; Indicates the Whether the candidate control item is activated, a value of 1 indicates activation, and a value of 0 indicates inactivation; represents the number of variables in the set of binary decision variables; Indicates the The time required for a candidate control item to take effect from startup; Indicates the penalty weight of dynamic environmental changes, and adjusts the weight of the impact of the environmental parameter change rate (such as smoke diffusion speed) on timeliness; It represents the gradient norm of environmental characteristics, integrating the intensity of dynamic parameters such as smoke concentration gradient and temperature change rate; represents the cost minimization objective; Indicates the The direct resource input costs caused by the candidate control items; Represents the opportunity cost weight, which adjusts the priority of indirect economic losses in the overall economic goal; Indicates the indirect economic losses caused by the candidate control items; represents the goal of minimizing risk; Represents the global risk adjustment coefficient, which amplifies or suppresses the impact of the event risk level on the security of the strategy; Indicates the event risk level, which is divided into risk levels according to the event type and severity, such as Level I low risk and Level III high risk; Indicates the The complex risk coefficient of candidate control items, the larger the value, the higher the risk; Indicates the A set of mutually exclusive controls. Controls in the same set cannot be activated at the same time. Represents a group of mutually exclusive control items; Indicates candidate control items is a candidate control Preconditions for activating candidate controls Candidate controls must be activated first ; Represents a pair of dependent controls; represents the constraint coefficient matrix, which describes the relationship between the control items and the tunnel structure parameters, such as lane width and ventilation shaft spacing; Represents the constraint threshold vector, which characterizes the physical limit value of the tunnel structure, such as maximum load and minimum clear height.

[0095] The improvements of the improved NSGA-III algorithm in the present invention include: guiding the search direction through a dynamic reference point and adaptively dividing the hyperplane according to the dimensions of the candidate control items; and adopting a constraint violation weighting strategy to penalize solutions that do not meet the tunnel structure constraints.

[0096] In some examples, the expression for guiding the search direction through the dynamic reference point is:

[0097] ;

[0098] in, Indicates a dynamic reference point; represents the minimum value in the minimization time objective; Represents the maximum value among the time objectives to be minimized; Represents the real-time environment offset that minimizes the time target, such as the fire spread stage Increase to favor time optimization; represents the minimum value in the cost minimization objective; represents the maximum value among the cost minimization objectives; represents the real-time environment offset that minimizes the cost objective; represents the minimum value in the risk minimization objective; represents the maximum value among the risk minimization objectives; Represents the real-time environment offset of minimizing the risk objective.

[0099] The present invention divides the target space hyperplane by dynamic reference points, thus avoiding the aggregation of solution sets caused by fixed reference points in traditional algorithms.

[0100] In some examples, a constraint violation weighting strategy is used to penalize solutions that do not satisfy the tunnel structure constraints. The expression is:

[0101] ;

[0102] in, Represents the total penalty value of constraint violation, which is a weighted penalty imposed on solutions that violate tunnel structure constraints. A larger value indicates a less feasible solution. Represents a dynamic adjustment coefficient, which adjusts the penalty intensity according to the urgency or security level of the incident. The larger the value, the stricter the penalty. represents the number of tunnel structure constraints; Indicates the Structural constraint function, quantitative solution The degree of violation of the tunnel's physical constraints, with positive values ​​indicating a violation.

[0103] In some examples, the interactive decision support interface outputs a final emergency strategy by calculating the minimum Manhattan distance solution from the Pareto optimal strategy set as the recommended strategy. The present invention uses Manhattan distance to emphasize the balance between objectives, avoiding extreme optimization of a single objective. Compared to Euclidean distance, this approach can better meet the trade-off requirements of emergency decision-making.

[0104] In some examples, the present invention can also dynamically update the feature extraction rule base and the weight parameters of the multi-objective optimization according to the execution results of the final emergency strategy; wherein, the dynamic update of the feature extraction rule base adopts an incremental reinforcement learning mechanism.

[0105] Example 2: A system for adaptively generating emergency strategies for sudden incidents in highway tunnels. The system is used to implement the method for adaptively generating emergency strategies for sudden incidents in highway tunnels as described in Example 1. Figure 2 As shown, it includes a data acquisition module, a feature extraction module, a control screening module and a target optimization module.

[0106] Among them, the data acquisition module is used to collect sensor data, video surveillance data and environmental parameters in the target tunnel to obtain multimodal data; the feature extraction module is used to extract the basic event features, tunnel structure features and environmental dynamic features layer by layer based on the multimodal data through the dynamic feature hierarchical extraction module to generate a multidimensional feature vector; the control screening module is used to perform primary screening of event types, secondary screening of severity and environmental adaptation verification in sequence based on the multidimensional feature vector through the control item dynamic screening module to screen out candidate control items from the total control item set; the target optimization module is used to perform multi-objective optimization of candidate control items using the NSGA-III algorithm or the improved NSGA-III algorithm to generate a Pareto optimal strategy set, and output the final emergency strategy in combination with the interactive decision support interface.

[0107] Working principle: The present invention constructs a multi-dimensional feature vector through a dynamic feature hierarchical extraction module, and combines a three-level feature fusion mechanism of event basic feature layer, tunnel structure feature layer and environmental dynamic feature layer. Compared with the traditional single-layer feature extraction method, it can effectively improve the accuracy of event type recognition and improve the timeliness of environmental risk prediction. Through the three-stage verification of the control item dynamic screening module, the screening accuracy of candidate control items is higher, effectively avoiding policy conflicts and resource waste.

[0108] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware. Furthermore, the present application may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0109] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0110] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0111] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0112] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method 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 in the scope of protection of the present invention.

Claims

1. A method for adaptively generating emergency strategies for sudden incidents in highway tunnels, characterized in that: The following steps are involved: Collect sensor data, video surveillance data, and environmental parameters in the target tunnel to obtain multimodal data; Based on the multimodal data, a dynamic feature hierarchical extraction module is used to extract event basic features, tunnel structure features, and environmental dynamic features layer by layer to generate a multidimensional feature vector; Based on the multidimensional feature vector, the control item dynamic screening module sequentially performs event type primary screening, severity secondary screening, and environmental adaptation verification to screen out candidate control items from the total control item set; Using the NSGA-III algorithm or the improved NSGA-III algorithm to perform multi-objective optimization on the candidate control items, generate a Pareto optimal strategy set, and output a final emergency strategy in combination with an interactive decision support interface; The event type initial screening is achieved using a multimodal fusion classification model; The modal fusion classification model takes the event basic features and the tunnel structure features as input, performs feature splicing operation on the event basic features and the tunnel structure features, and then applies a multi-classification activation function to output a probability distribution vector of the event type; The secondary screening of severity is achieved through risk index calculation, specifically including: Solving the environmental dynamic characteristics inversely in time to obtain the environmental change risk; Normalizing the basic features of the event and calculating the L2 norm to obtain the global event risk; Determining a risk index based on the sum of the environmental change risk and the environmental change risk; The environment adaptation verification includes: Perform fluid dynamics simulation verification on ventilation system controls; Perform crowd dynamics model validation on evacuation path planning items; And / or, perform reachability topology analysis on fire rescue deployment items.

2. A method for adaptively generating emergency strategies for highway tunnel emergencies according to claim 1, characterized in that: The dynamic feature hierarchical extraction module includes: The event basic feature layer is used to extract event basic features including temperature, smoke concentration, CO concentration, vehicle density and accident point location data; Tunnel structural feature layer, used to extract tunnel structural features including tunnel curvature, ventilation shaft distribution, escape route topology, and fire protection facility layout; The environmental dynamic feature layer is used to extract environmental dynamic features including wind speed vector field, temperature and humidity gradient matrix and visibility attenuation rate.

3. The method for adaptively generating emergency strategies for highway tunnel emergencies according to claim 1, characterized in that: The objective function of the multi-objective optimization uses the activation state of the candidate control item as a decision variable; The objective function includes: Minimize the time objective, including the coupled effects of control item execution time and dynamic environment gradients; Minimize cost objectives and introduce opportunity costs to quantify the losses caused by abandoning other measures due to enabling control items; Minimize the risk objective and use the product form to represent the complex risk amplification effect brought about by the superposition of multiple control items; Constraints include mutually exclusive control item constraints, dependent control item constraints, and tunnel physical constraints.

4. The method for adaptively generating emergency strategies for highway tunnel emergencies according to claim 1, characterized in that: The improvements of the improved NSGA-III algorithm include: Guiding the search direction through a dynamic reference point and adaptively dividing the hyperplane according to the dimensions of the candidate control items; Furthermore, a constraint violation weighting strategy is adopted to penalize solutions that do not satisfy the tunnel structure constraints.

5. The method for adaptively generating emergency strategies for highway tunnel emergencies according to claim 1, characterized in that: The interactive decision support interface outputs a final emergency strategy, including: The minimum Manhattan distance solution is calculated in the Pareto optimal strategy set as the recommended strategy, and the final emergency strategy is output.

6. The method for adaptively generating emergency strategies for highway tunnel emergencies according to claim 1, characterized in that: The method further includes: Dynamically updating a feature extraction rule base and weight parameters of the multi-objective optimization according to the execution results of the final emergency strategy; The dynamically updated feature extraction rule base adopts an incremental reinforcement learning mechanism.

7. A system for adaptively generating emergency strategies for sudden incidents in highway tunnels, characterized in that: The system is used to implement a method for adaptively generating emergency strategies for sudden incidents in highway tunnels as described in any one of claims 1 to 6, comprising: The data acquisition module is used to collect sensor data, video surveillance data and environmental parameters in the target tunnel to obtain multimodal data; A feature extraction module is used to extract event basic features, tunnel structure features and environmental dynamic features layer by layer based on the multimodal data through a dynamic feature hierarchical extraction module to generate a multidimensional feature vector; A control screening module is used to perform event type primary screening, severity secondary screening and environment adaptation verification in sequence through a control item dynamic screening module based on the multi-dimensional feature vector, and screen out candidate control items from the total control item set; The target optimization module is used to perform multi-objective optimization on the candidate control items using the NSGA-III algorithm or the improved NSGA-III algorithm, generate a Pareto optimal strategy set, and output the final emergency strategy in combination with an interactive decision support interface.

Images