Hidden access emergency response intelligent collaborative awareness and decision-making system

Abstract

The invention relates to the technical field of traffic scheduling, and particularly discloses a hidden channel emergency response intelligent collaborative perception and decision-making system, which comprises a multi-level heterogeneous module, a data linkage query module, a channel parameter prediction module, a traffic state evaluation module, an emergency defibering scheduling module and an information playback and display module. According to the method, through the functions of rapid query, such as keyword retrieval, advanced retrieval of complex query conditions, one-stop full-text retrieval and batch retrieval, it is ensured that workers of different departments and levels can efficiently obtain needed information, circulation and sharing of information can be promoted, and the efficiency of information query is improved. All related departments can make decisions based on newest and comprehensive information, the information island phenomenon is reduced, the overall emergency rescue efficiency is improved, and the whole process of recording and return visit of event awareness-comprehensive decision-rescue guarantee can be achieved based on a comprehensive geographic information system (GIS).

Description

Technical Field

[0001] The present invention belongs to the technical field of traffic dispatching, and particularly relates to an intelligent collaborative perception and decision-making system for emergency response of the Tibet-inbound channels. Background Art

[0002] The Tibet-inbound channels are located in extreme environmental conditions such as high cold, high altitude, and large temperature differences, and are often threatened by natural disasters such as avalanches, landslides, and road icing. These disasters not only have a serious impact on traffic, but also increase the frequency of traffic accidents, and may cause casualties and interruption of the transportation of important materials, which puts great pressure on the emergency support system. To address the above challenges, a set of efficient disaster monitoring, early warning, and response mechanisms has been established: the "event perception - comprehensive decision-making - rescue support" display technology for the Tibet-inbound channels in emergency scenarios.

[0003] The special environment and complex disaster situation of the Tibet-inbound channels require accurate event perception, comprehensive decision-making, and efficient rescue support. This technology uses advanced technical means - deep learning and intelligent perception technology - to timely identify potential disaster risks and predict the probability of disasters, laying a foundation for rapid response. At the same time, the Tibet-inbound channels involve the cooperation of multiple departments such as transportation, emergency, meteorology, and security. Efficient information integration and sharing are required among these departments to ensure the implementation of rapid decision-making and emergency response.

[0004] Therefore, this technology realizes cross-departmental and cross-hierarchical collaborative work by integrating various data sources, performing data linkage, rapid retrieval, and comprehensive display, effectively improving the efficiency of disaster response. This technology can not only efficiently process heterogeneous data such as geographic information, meteorological data, and traffic flow data, but also support the decision-making layer to make accurate decisions through real-time information and dynamic analysis, optimize resource allocation, and ensure the timeliness of emergency response and rescue efficiency. During a disaster, this technology provides strong support for emergency decision-making and resource dispatching through the integration of multiple technologies and resources, helps to quickly restore traffic, and ensures road safety.

[0005] In view of this, the inventor proposes an intelligent collaborative perception and decision-making system for emergency response of the Tibet-inbound channels to solve the above problems. Summary of the Invention

[0006] The purpose of the present invention is to provide an intelligent collaborative perception and decision-making system for emergency response of the Tibet-inbound channels to solve the problems raised in the above background art.

[0007] To achieve the above purpose, the present invention provides the following technical solutions: An intelligent collaborative perception and decision-making system for emergency response of the Tibet-inbound channels, comprising: Multi-level heterogeneous modules: used to integrate data from different departments and levels to obtain a heterogeneous data set for unified management and invocation; Data linkage query module: used to retrieve the heterogeneous data set, obtain key data, extract relevant information from the key data according to user needs, and generate a structured query result; Channel parameter prediction module: used to deeply analyze the query result, predict road traffic capacity and potential risks based on spatio-temporal dependence modeling, parameter sharing strategy, and multi-task learning model, and generate traffic change prediction data at the node level and section level; Traffic status evaluation module: used to perform unsupervised clustering and adaptive evaluation on the traffic change prediction data by using fuzzy clustering algorithm and dynamic graph convolutional neural network model, output the traffic operation status, and obtain the operation status evaluation report of each section; Emergency evacuation scheduling module: according to the operation status evaluation report, adopt dynamic path selection and multi-objective optimization model to formulate the optimal evacuation and vehicle scheduling plan, and obtain the dynamic evacuation and resource allocation plan; Information playback and display module: used to perform dynamic visualization display on the dynamic evacuation and resource allocation plan based on GIS technology.

[0008] Preferably, the spatio-temporal dependence modeling encodes traffic observation data into traffic volume based on nodes and traffic volume based on sections respectively, constructs a node graph and a section graph to reflect the dual hierarchical structure of the road network; Using the dynamic graph self-attention mechanism, respectively learn the potential semantics of traffic changes in the node graph and the section graph extract spatio-temporal features to capture the dynamic influence pattern of abnormal events.

[0009] Preferably, the parameter sharing strategy considers the potential correlation between the node graph and the section graph, designs a parameter sharing strategy, embeds shared parameters in the two graph structures, and learns the common information between the two through a common non-linear transformation formula: ; where, and are shared weight parameters; Use the multi-task fusion module to output the node-level prediction result The hidden correlation between these sections and intersections can be coupled through multi-task learning.

[0010] Preferably, the multi-task learning model fuses node and section features at the feature level, uses the multi-task learning module to jointly output traffic change prediction results at the node level and section level, and captures the hidden mutual enhancement relationship between nodes and sections through multi-task coupled learning.​

[0011] Preferably, the fuzzy clustering algorithm is used to solve the problem of traffic operation state evaluation, including three parts: objective function optimization, conditional constraints, and process optimization; The function optimization uses the weighted sum of squared class errors as the optimization objective function for clustering : ; Among them, is the sample data, and its element is the th sample, is the clustering center matrix, and the element is the th clustering center point, represents the relationship value between the th sample and the th clustering center point. The values of each element form the membership matrix , is the number of clustering categories, is the number of samples, is the Euclidean distance, is a hyperparameter that reflects the weighting effect.

[0012] Preferably, the constraint condition expression for minimizing the objective function in the conditional constraints is: ; The sum of the relationships between each sample point and the clustering center is 1, and there are constraint relationships for sample points.

[0013] Preferably, the process optimization is obtained from the parameter update of the objective function optimization and the conditional constraints. The parameter update formula for the process optimization includes: Update the membership matrix: ; Update the clustering center: ; Preferably, the steps of the dynamic path selection are as follows: S1. The traffic network is represented by a graph. Therefore, based on the road network structure, a road network model is established as , where N is the node set, representing the set of each intersection in the road network, E is the link set, representing the connection relationship between each intersection, i.e., the road section, and W is the link weight set, representing the weight vector of the link. Different weights can be assigned to the road sections. represents the set of safe points s, I represents the set of road sections i, and the road section i leading to the safe point s is ; S2. In the road network model, the traffic capacity and length of each road segment determine its weight. By adjusting the weight ratio of the traffic capacity and length of the road segment under different demand scenarios through the weight coefficient, the shortest and high-traffic-capacity route is selected. The Dijkstra algorithm is used to calculate the shortest path, and the specific steps are as follows: S3. Initialization: For each vertex in the graph, set the estimated shortest distance from the starting point to this vertex to infinity, except that the distance from the starting point to itself is 0. Maintain a priority queue for storing all vertices and their current shortest path estimates to quickly select the next vertex to be processed; S4. Update distance: Starting from the starting point, update the minimum weight estimates of all its adjacent vertices. For each vertex adjacent to the starting point, if the total weight to reach it through the current vertex is smaller than the known minimum weight, update the minimum weight estimate of this vertex; S5. Select vertex: Select the vertex with the smallest current minimum weight estimate from the priority queue as the new current vertex; S6. Repeat steps S2 and S3: For the new current vertex, repeat the process of updating the total path weight and selecting the vertex in steps S2 and S3 until all vertices have been processed or the priority queue is empty; S7. Construct the relative shortest path: Use the precursor node information to backtrack the relative shortest path from the evacuation starting point to the s safe point ; The set of road segments for all route trips between the evacuation point and each safe point is stored according to the flow direction of the traffic flow. All total routes are L, and the road segment to the s safe point is , .

[0014] Preferably, the expression of the multi-objective optimization model is: ; The specific constraint conditions are as follows: ; Where is the traffic flow prediction information of the road segment, which will be continuously updated as the prediction information is updated, so as to achieve dynamic evacuation scheduling; T is the total evacuation time, indicating the overall time for the evacuation vehicles to reach each safe point from the starting point. The objective function aims to minimize this parameter; D is the vehicle travel distance, indicating the total distance of the vehicle travel path or the road segment length, which jointly affects the evacuation efficiency with the travel time; V is the number of vehicles, indicating the total number of vehicles to be evacuated or the number of vehicles allocated to each safe point s; α and β are weight coefficients used to balance the priorities among multiple objectives. α is used to weight the objective of minimizing time, and β is used to weight the objective of the reliability of traffic capacity. ns is the number of safety points, representing the total number of safety points in the system and used as an index in the constraint conditions. Ks is the capacity of safety point s, representing the maximum number of vehicles that safety point s can accommodate. fs is the traffic capacity of section s, representing the number of vehicles that can pass through section s per unit time. c is the environmental adjustment coefficient, used to dynamically adjust the relationship between the traffic capacity of the section and the number of allocated vehicles, and reflecting the attenuation effect of environmental factors (such as weather, disasters) on the traffic capacity.

[0015] Preferably, the data of different departments and levels include accident disaster information, emergency support information, road network operation status information, and emergency rescue material information. The retrieval includes quick retrieval through keywords, complex condition retrieval, and associated query.

[0016] Compared with the prior art, the beneficial effects of the present invention are as follows: (1) Through functions such as quick query, such as keyword retrieval, advanced retrieval with complex query conditions, one-stop full-text retrieval, batch retrieval, etc., the present invention ensures that staff at different departments and levels can efficiently obtain the required information. The present invention can promote the circulation and sharing of information, ensure that all relevant departments can make decisions based on the latest and comprehensive information, reduce the phenomenon of information islands, and improve the overall emergency rescue efficiency.

[0017] (2) Based on the integrated Geographic Information System (GIS), the present invention can record and review the whole process of "event perception - comprehensive decision-making - rescue support". The present invention can realize multi-dimensional and comprehensive display of information such as accident disaster information, channel operation status, emergency support information, and rescue materials. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 It is a block diagram of the composition of an intelligent collaborative perception and decision-making system for emergency response of the Tibet-bound channel of the present invention; Figure 2 It is the overall structure diagram of the emergency evacuation and dispatching of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0019] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0020] Example 1:

[0021] Please refer to Figure 1 and Figure 2 as shown, an intelligent collaborative perception and decision-making system for the emergency response of the Tibet-bound channel, including: Multi-level heterogeneous module: used to integrate data from different departments and levels to obtain a heterogeneous data set for unified management and invocation; Data linkage query module: used to retrieve the heterogeneous data set, obtain key data, extract relevant information from the key data according to user needs, and generate a structured query result; Channel parameter prediction module: used to deeply analyze the query result, predict the road traffic capacity and potential risks based on spatio-temporal dependence modeling, parameter sharing strategy, and multi-task learning model, and generate traffic change prediction data at the node level and section level; Traffic state evaluation module: used to perform unsupervised clustering and adaptive evaluation on the traffic change prediction data by using the fuzzy clustering algorithm and the dynamic graph convolutional neural network model, output the traffic operation state, and obtain the operation state evaluation report of each section.

[0022] Emergency evacuation scheduling module: according to the operation state evaluation report, adopt the dynamic path selection and multi-objective optimization model to formulate the optimal evacuation and vehicle scheduling plan, and obtain the dynamic evacuation and resource allocation plan; Information playback and display module: used to dynamically visualize the dynamic evacuation and resource allocation plan based on GIS technology, and support the whole-process playback and multimedia integrated display, providing an intuitive display interface for command and decision-making, including the whole-process backtracking and dynamic adjustment of disaster occurrence, response, evacuation, and recovery; Specifically, the spatio-temporal dependence modeling encodes traffic observation data into traffic volume based on nodes and traffic volume based on sections to respectively construct a node graph and a section graph to reflect the dual hierarchical structure of the road network; Using the dynamic graph self-attention mechanism, respectively learn the potential semantics of traffic changes in the node graph and the section graph to extract spatio-temporal features to capture the dynamic influence pattern of abnormal events.

[0023] Specifically, the parameter sharing strategy considers the potential correlation between the node graph and the section graph, designs a parameter sharing strategy, embeds shared parameters in the two graph structures, and learns their common information through a common non-linear transformation formula: ; The common features and 、 Output node-level prediction results using the multi-task fusion module The hidden correlations between these road segments and intersections can be coupled through multi-task learning to enhance the prediction performance mutually.

[0024] Specifically, the multi-task learning model fuses node and road segment features at the feature level, and uses the multi-task learning module to jointly output traffic change prediction results at the node level and road segment level. Through multi-task coupling learning, the hidden mutual enhancement relationship between nodes and road segments is captured to improve the overall prediction performance.

[0025] Specifically, the fuzzy clustering algorithm is used to solve the problem of traffic operation state evaluation, including three parts: objective function optimization, conditional constraints, and process optimization; The function optimization uses the weighted sum of squared class errors as the optimization objective function for clustering : ; Among them, is the sample data, and its element is the th sample, is the cluster center matrix, and the element is the th cluster center point, represents the relationship value between the th sample and the th cluster center point. The values of each element form the membership matrix , is the number of clustering categories, is the number of samples, is the Euclidean distance, is a hyperparameter that reflects the effect of weighting.

[0026] Specifically, the constraint condition expression for minimizing the objective function in the conditional constraints is: ; The sum of the relationships between each sample point and the cluster center is 1, and there are constraint relationships for sample points.

[0027] Specifically, the process optimization is obtained from the parameter updates of the objective function optimization and conditional constraints. The parameter update formulas for the process optimization include: Update the membership matrix: ; Update the cluster center: .

[0028] Specifically, the steps of the dynamic path selection are as follows: S1. The traffic network is represented by a graph. Therefore, based on the road network structure, a road network model is established as , where N is the set of nodes, representing the set of each intersection in the road network, E is the set of links, representing the connection relationship between each intersection, that is, the road section, and W is the set of link weights, representing the weight vector of the link, and different weights can be assigned to the road sections. represents the set of safe points s, I represents the set of road sections i, and the road section i leading to the safe point s is ; S2. In the road network model, the traffic capacity and length of each road section determine its weight. By adjusting the weight ratio of the traffic capacity and length of the road section under different demand scenarios through the weight coefficient, the shortest and highest traffic capacity route is selected. The Dijkstra algorithm is used to calculate the shortest path, and the specific steps are as follows: S3. Initialization: For each vertex in the graph, set the estimated shortest distance from the starting point to this vertex to infinity, except that the distance from the starting point to itself is 0; maintain a priority queue for storing all vertices and their current shortest path estimates to quickly select the next vertex to be processed; S4. Update the distance: Starting from the starting point, update the minimum weight estimate of all its adjacent vertices. For each vertex adjacent to the starting point, if the total weight to reach it through the current vertex is smaller than the known minimum weight, update the minimum weight estimate of this vertex; S5. Select a vertex: Select the vertex with the smallest current minimum weight estimate value from the priority queue as the new current vertex; S6. Repeat steps S2 and S3: For the new current vertex, repeat the process of updating the total path weight and selecting a vertex in steps S2 and S3 until all vertices have been processed or the priority queue is empty; S7. Construct the relative shortest path: Use the information of the predecessor node to backtrack the relative shortest path from the evacuation starting point to the safe point s ; The set of road sections of all route trips between the evacuation point and each safe point is stored according to the flow direction of the traffic flow. All the total routes are L, and the road section leading to the safe point s is .

[0029] Specifically, the expression of the multi-objective optimization model is: It is necessary to dynamically adjust the vehicle scheduling and allocation according to the traffic flow prediction data, the traffic capacity of the road section, and the vehicle evacuation demand. The goal of this scheme is to ensure the timeliness and safety of evacuation, and to minimize the evacuation time and ensure the reliability of the traffic capacity of the road section by optimizing the path selection, the number of dispatched vehicles, and controlling road congestion.

[0030] For this purpose, a multi-objective optimization model is established as follows: The objective function includes minimizing the evacuation time and maximizing the reliability of the road section capacity; ; The specific constraint conditions are as follows: ; Among them, is the traffic flow prediction information of the road section, which will be continuously updated as the prediction information is updated, so as to achieve dynamic evacuation scheduling; T is the total evacuation time, which represents the overall time for the evacuated vehicles to reach each safe point from the starting point. The objective function aims to minimize this parameter; D is the vehicle travel distance, which represents the total distance of the vehicle travel path or the road section length, and jointly affects the evacuation efficiency with the travel time; V is the number of vehicles, which represents the total number of vehicles to be evacuated, or the number of vehicles assigned to each safe point s; α and β are weight coefficients, which are used to balance the priorities among multiple objectives. α is used to weight the objective of minimizing time, and β is used to weight the objective of the reliability of the road section capacity; ns is the number of safe points, which represents the total number of safe points in the system and is used as an index in the constraint conditions; Ks is the capacity of safe point s, which represents the maximum number of vehicles that safe point s can accommodate; fs is the capacity of road section s, which represents the number of vehicles that can pass through road section s per unit time; c is the environmental adjustment coefficient, which is used to dynamically adjust the relationship between the road section capacity and the number of assigned vehicles, and reflects the attenuation effect of environmental factors (such as weather, disasters) on the road section capacity; In the above formula, the objective function of formula (1) aims to minimize the overall time for the evacuated vehicles to reach each safe point.

[0031] The objective function of formula (2) ensures the evacuation speed through a road section capacity reliability function, and avoids secondary accidents caused by traffic congestion on the evacuation roads.

[0032] In formula (3), it is ensured that the road section capacity to reach safe point s is greater than or equal to the number of assigned vehicles, and c is the environmental adjustment coefficient.

[0033] In formula (4), all parameters are greater than 0.

[0034] In formula (5), it is ensured that the number of vehicles assigned to each safe point is equal to the number of vehicles to be evacuated and scheduled.

[0035] In formula (6), it is ensured that the number of vehicles at each safe point does not exceed the maximum capacity.

[0036] In formula (7), ensure the path consistency of vehicle allocation, which is consistent with the shortest path.

[0037] To solve this multi-objective optimization model, this solution uses an improved NSGA-2 algorithm. This algorithm includes the following steps when solving problems: Coding operation: Use the gene sequence coding method to represent the vehicle scheduling and allocation to ensure consistency with the optimization objectives.

[0038] Initializing the population: Initialize the population by the random method and initialize it according to the scheduling results of the previous round to accelerate the iteration speed.

[0039] Non-dominated sorting: Sort the solutions according to the dominance relationship between the solutions to obtain the Pareto optimal solutions.

[0040] Crowding distance calculation: Calculate the crowding distance of individuals to maintain the diversity of the population and avoid localization of the solution space.

[0041] Selection operation: Adopt the tournament selection method to select individuals with better fitness to enter the next generation population.

[0042] Dynamically adjust the crossover and mutation rates: Dynamically adjust the crossover rate and mutation rate according to different stages of the algorithm iteration to optimize the search process.

[0043] Population merging and pruning: Merge and prune the parent and offspring populations, and retain individuals with higher diversity.

[0044] Specifically, the data of different departments and levels include accident disaster information, emergency support information, road network operation status information, and emergency rescue material information; The retrieval includes quick retrieval through keyword retrieval, complex condition retrieval, and association query.

[0045] As can be seen from the above, the present invention ensures that staff at different departments and levels can efficiently obtain the required information through functions such as quick query, such as keyword retrieval, advanced retrieval with complex query conditions, one-stop full-text retrieval, and batch retrieval.

[0046] Optimize emergency response and decision support: The present invention can promote the circulation and sharing of information, ensure that all relevant departments can make decisions based on the latest and comprehensive information, reduce the information island phenomenon, and improve the overall emergency rescue efficiency.

[0047] Implement full-process recording and playback: Based on the integrated Geographic Information System (GIS), the present invention can record and playback the entire process of "event perception - comprehensive decision-making - rescue support".

[0048] Multi-level information display: The present invention can realize multi-dimensional and comprehensive display of accident and disaster information, channel operation status, emergency support information, rescue materials and other information.

[0049] Embodiment 2: The present invention will explain the specific functions of the display technology in order to more intuitively demonstrate its advantages. Its specific functions include: home page overview, data detection function display, emergency dispatch function display and history query function display.

[0050] Further, home page overview: Display the real-time operating status of the four channels into Tibet, including traffic flow, traffic status and blocking events.

[0051] Click on the observation station location on the route to view the observation station details.

[0052] The management and maintenance units are responsible for monitoring and maintaining the four channels into Tibet. The platform functions include identification and display of the management and maintenance units.

[0053] Data detection function display: The platform can check the current congestion situation of the road network and link the map to display the location of congested sections and jurisdiction information; The platform integrates and displays detailed information about blocking events, including route number, name, starting and ending stake numbers, and geographic coordinates. In addition, the system can also distinguish event types (such as blockage or interruption). Users can grasp event dynamics in real time and analyze blocking trends.

[0054] The platform can identify specific sections or areas within the management and maintenance area that are prone to disasters or traffic accidents due to factors such as geography, climate, and traffic flow.

[0055] Emergency dispatch function display: The platform provides an interactive scheduling interface that supports event classification, resource calling, and dynamic simulation display.

[0056] History query function display: The platform provides query and export functions, supports trend analysis of historical data, and helps users review data and make future forecasts.

[0057] As can be seen from the above, the purpose of the "Event Perception-Comprehensive Decision-Rescue Guarantee" demonstration technology for the Tibet Access Channel in emergency scenarios is to improve the efficiency and accuracy of emergency response. By integrating deep learning, intelligent perception and data linkage technology, it can timely identify potential disaster risks, predict the probability of disasters, and support rapid decision-making and rescue guarantees. It aims to achieve cross-departmental collaboration and resource optimization, ensure that traffic can be quickly restored when a disaster occurs, reduce casualties and interruptions in material transportation, and improve the disaster response capabilities and emergency guarantee level of the Tibet Access Channel.

[0058] Example 3: Emergency Response to Avalanche Blockage Incidents: An avalanche occurred on a certain section of the Tibet-bound route (section number A-12), resulting in a complete blockage of two-way traffic. The snow cover thickness around the section was 1.5 meters, the road length was 500 meters, and the snow accumulation continued to increase.

[0059] Input Data: Meteorological Data: Snow cover thickness 1.5 m, temperature -10°C, wind speed 25 km / h; Road Network Information: Section number A-12, length 500 m, surrounding slope 20°; Traffic Flow: The average daily traffic volume of this section is 4000 vehicles, and the current number of stranded vehicles is 600. Initial Conditions: Snow cover thickness H0 = 1.5 m, snow accumulation rate Rs = 0.1 m / h.

[0060] Traffic Flow: Average traffic volume Q0 = 4000 vehicles / day, stranded vehicles N0 = 600 vehicles.

[0061] Road Information: Section length L = 500 m.

[0062] Rescue Vehicle Capacity: Each evacuation Cv = 50 vehicles.

[0063] Result: Integrate the above heterogeneous data to form a structured data set DA−12.

[0064] Data Linkage Query Module Extracts Key Information Conditional Query: Geographical Location: Section A-12.

[0065] Event Type: Avalanche Blockage.

[0066] Result: Output the snow depth, number of stranded vehicles, and related weather forecasts for section A-12.

[0067] Channel Parameter Prediction Module: Prediction Method: Based on a dynamic graph neural network model (time step is 10 minutes).

[0068] Prediction Result: The snow cover thickness will increase to 2.0 meters within the next hour.

[0069] The number of stranded vehicles will rise to 800.

[0070] Traffic Status Evaluation Module: Fuzzy Clustering Result: The current status of this section is "severely congested".

[0071] Adaptive Evaluation Result: Traffic recovery is expected to take 8 hours (under the condition of no human intervention).

[0072] Emergency evacuation scheduling module: Route selection: Use the Dijkstra algorithm to plan a detour route: A-11 → A-13.

[0073] Evacuation vehicles: Based on the NSGA-2 optimization algorithm, dispatch 20 rescue vehicles and evacuate 50 vehicles at a time.

[0074] Information playback and display module: GIS display: Real-time display of the avalanche impact area and rescue progress.

[0075] Multimedia integration: Photos of snow cover and dynamic videos of evacuation routes.

[0076] Snow accumulation growth prediction Ht = H0 + Rs⋅t Predict the snow depth in the next 1 hour: H1 = 1.5 + 0.1⋅1 = 1.6 m Stranded vehicle growth prediction: Assume the ratio of stranded vehicle growth to traffic flow reduction is Rv = 0.1 vehicle / min: Nt = N0 + Rv⋅t Predict the number of stranded vehicles within 1 hour: N1 = 600 + 0.1⋅60 = 660 vehicles Rescue vehicle evacuation time The cleaning rate of a single rescue vehicle is Rc = 50 m / h, and 500 meters need to be cleared: Tclear = L / Rc = 500 / 50 = 10 h 20 rescue vehicles work simultaneously: Tclearopt = Tclear / 20 = 10 / 20 = 0.5 h Evacuation vehicle optimization The number of vehicles evacuated each time is Cv = 50: Tevac = N1 / Cv = 660 / 50 = 13.2 rounds If each round takes 20 minutes, the total evacuation time: Ttotal = Tevac×20 min = 13.2×20 = 264 min = 4.4 h As can be seen from the above, the predicted data: The snow depth will increase to 1.6 meters in the next 1 hour, and the number of stranded vehicles will increase to 660.

[0077] Rescue effect: Use 20 rescue vehicles, the cleaning time is 0.5 hours, and the evacuation is completed within 4.4 hours.

[0078] Thus, the road traffic capacity can be quickly restored, and the impact of traffic congestion can be reduced.

[0079] Example 4 Debris flow secondary disaster warning and response: Due to continuous heavy rainfall in a mountainous section (numbered B-05), the river water level rose by 20%, resulting in an increased risk of debris flow.

[0080] Multi-level heterogeneous module integrates data: Input data: Rainfall: The cumulative rainfall in the past 24 hours was 120 mm, and continuous rainfall is expected for 48 hours.

[0081] Hydrological data: The river water level rose by 20% compared to the normal value.

[0082] Section information: B-05, the road surface is slippery, and 30% is covered with sediment.

[0083] Result: Generate the comprehensive data set DB−05 The data linkage query module extracts key information: Conditional query: Rainfall > 100 mm.

[0084] River water level > 20% of the normal value.

[0085] Result: The debris flow risk level of section B-05 is evaluated as "high risk".

[0086] Channel parameter prediction module: Prediction method: Multi-task learning model to predict the debris flow triggering probability.

[0087] Prediction result: The debris flow triggering probability within the next 12 hours is 85%.

[0088] Traffic status evaluation module Fuzzy clustering result: The section status is "potentially closed".

[0089] Adaptive evaluation result: It is recommended to evacuate vehicles in advance and close the section.

[0090] Emergency evacuation and dispatching module: Route selection: Plan the detour route B-04 → B-06, with an estimated additional travel time of 30 minutes.

[0091] Dispatching plan: Evacuate 200 vehicles in batches and dispatch 3 dredging equipment.

[0092] Information playback and display module: GIS display: Dynamically display the water level changes and risk areas.

[0093] Multimedia integration: rainfall distribution map, real-time monitoring video; Initial conditions: cumulative rainfall R0 = 120 mm, rainfall growth rate Gr = 5 mm / h.

[0094] River water level: normal value is 10 m, current water level is 12 m (a 20% increase).

[0095] Evacuation vehicle demand: total number of vehicles V0 = 200 vehicles.

[0096] Rainfall prediction: Rt = R0 + Gr⋅t Predicted rainfall in the next 6 hours: R6 = 120 + 5⋅6 = 150 mm Formula for predicting the probability P of debris flow triggering: P = 1−e −λ⋅Rt Assume the trigger sensitivity coefficient λ = 0.02: P = 1−e −0.02⋅150 ≈0.95 Traffic status evaluation Using the fuzzy clustering method, the traffic status is divided into four categories (smooth, slow, congested, closed). If the debris flow trigger probability P > 0.8, the status is "closed".

[0097] Optimization of vehicle evacuation time The total length of the detour route is 30 km, the vehicle speed Vs = 30 km / h, detour time: Troute = 30 / 30 = 1 h Number of evacuated vehicles per batch Cv = 50: Tevac = V0 / Cv = 200 / 50 = 4 rounds Evacuation time per round is 15 minutes, total evacuation time: Ttotal = Tevac⋅15 min = 4×15 min = 60 min = 1 h As can be seen from the above, the predicted data: rainfall in the next 6 hours reaches 150 mm, and the debris flow trigger probability is 95%.

[0098] Rescue effect: early warning, vehicle evacuation completed within 1 hour, and effectively detoured to a safe route.

[0099] Avoid the impact of secondary disasters and ensure vehicle safety.

[0100] Although embodiments of the present invention have been shown and described, it will be understood by those of ordinary skill in the art that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the present invention, and the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. An intelligent collaborative perception and decision-making system for emergency response of the Tibet entry channel, characterized in that Including: Multi-level heterogeneous module: used to integrate data from different departments and levels to obtain a heterogeneous data set for unified management and invocation; Data linkage query module: used to retrieve the heterogeneous data set, obtain key data, extract relevant information from the key data according to user needs, and generate a structured query result; Channel parameter prediction module: used to deeply analyze the query result, predict road traffic capacity and potential risks based on spatio-temporal dependence modeling, parameter sharing strategy, and multi-task learning model, and generate traffic change prediction data at the node level and section level; Traffic state evaluation module: used to perform unsupervised clustering and adaptive evaluation on the traffic change prediction data by using fuzzy clustering algorithm and dynamic graph convolutional neural network model, output the traffic operation state, and obtain the operation state evaluation report of each section; Emergency evacuation scheduling module: according to the operation state evaluation report, adopt dynamic path selection and multi-objective optimization model to formulate the optimal evacuation and vehicle scheduling plan, and obtain the dynamic evacuation and resource allocation plan; Information playback and display module: used to perform dynamic visualization display on the dynamic evacuation and resource allocation plan based on integrated geographic information system technology.

2. The intelligent collaborative perception and decision-making system for emergency response of the Tibet-bound passage according to claim 1, wherein The spatio-temporal dependence modeling encodes traffic observation data as traffic volume based on nodes and traffic volume based on road segments to construct a node graph and a road segment graph respectively to reflect the dual hierarchical structure of the road network; Using the dynamic graph self-attention mechanism, learn the latent semantics of traffic changes in the node graph and the road segment graph respectively, and extract spatio-temporal features to capture the dynamic influence patterns of abnormal events.

3. The intelligent collaborative perception and decision-making system for emergency response of the Tibet-bound passage according to claim 2, wherein, The parameter sharing strategy considers the potential correlation between the node graph and the section graph, designs a parameter sharing strategy, embeds shared parameters in the two graph structures, and learns their common information through a common non-linear transformation formula: ； Among them, and are shared weight parameters; The shared features and 、 Use the multi-task fusion module to output node-level prediction results , and the hidden correlation between road segments and intersections can be coupled through multi-task learning.

4. The intelligent collaborative perception and decision-making system for emergency response of the Tibet access road according to claim 3, characterized in that, The multi-task learning model fuses node and section features at the feature level, uses a multi-task learning module to jointly output traffic change prediction results at the node level and section level, and captures the hidden mutual enhancement relationship between nodes and sections through multi-task coupling learning.

5. An intelligent collaborative perception and decision-making system for emergency response of the Tibet-bound passage according to claim 1, characterized in that, The fuzzy clustering algorithm is used to solve the traffic operation state evaluation problem, including three parts: objective function optimization, conditional constraints, and process optimization; The function optimization uses the weighted sum of squared class errors as the optimization objective function for clustering : ； Among them, is the sample data, and its element is the th sample, is the cluster center matrix, and the element is the th cluster center point, represents the relationship value between the th sample and the th cluster center point. The values of each element form the membership degree matrix , is the number of categories of clustering, is the number of samples, is the Euclidean distance, is a hyperparameter that reflects the effect of weighting.

6. The intelligent collaborative perception and decision-making system for emergency response of the Tibet access channel according to claim 5, characterized in that, The constraint condition expression for minimizing the objective function in the conditional constraints is: ； The sum of the relationships between each sample point and the cluster center is 1, There are constraint relationships for each sample point.

7. An intelligent collaborative perception and decision-making system for emergency response of the Tibet-bound passage according to claim 5, characterized in that, The process optimization is obtained from the parameter update of the objective function optimization and conditional constraints. The parameter update formula for the process optimization includes: Update the membership matrix: ； Update the cluster center: ； Among them, is the sample data, and its element is the th sample, is the cluster center matrix, and its element is the th cluster center point, represents the relationship value between the th sample and the th cluster center point. The values of each element form the membership degree matrix , is the number of cluster categories, is the number of samples, is the Euclidean distance, is a hyperparameter that reflects the weighting effect. weights the distance. The closer to the center, the larger the value; the farther from the center, the smaller the value. When is too large, the weighting effect decreases, and the distances tend to be the same; By minimizing the objective function, the optimal clustering criterion: membership matrix and clustering center matrix are obtained, thereby obtaining the clustering result.

8. An intelligent collaborative perception and decision-making system for emergency response of the Tibet-bound passage according to claim 1, characterized in that, The steps of the dynamic path selection are as follows: S1. The traffic network is represented by a graph. Therefore, according to the road network structure, a road network model is established as , where N is the set of nodes, representing the set of each intersection of the road network; E is the set of links, representing the connection relationship between each intersection, that is, the road section; W is the set of link weights, representing the weight vector of the link, and different weights are assigned to the road sections. represents the set of safe points s, I represents the set of road sections i, and the road section i leading to the safe point s is ; S2. In the road network model, the traffic capacity and length of each section determine its weight. By adjusting the weight ratio of the traffic capacity and length of the section under different demand scenarios through the weight coefficient, the shortest and high-traffic-capacity route is selected; the Dijkstra algorithm is used to calculate the shortest path, and the specific steps are as follows: S3. Initialization: For each vertex in the graph, set the estimated shortest distance from the starting point to this vertex to infinity, except that the distance from the starting point to itself is 0; maintain a priority queue for storing all vertices and their current shortest path estimated values to quickly select the next vertex to be processed; S4. Update the distance: Starting from the starting point, update the minimum weight estimate of all its adjacent vertices. For each vertex adjacent to the starting point, if the total weight to reach it through the current vertex is smaller than the known minimum weight, update the minimum weight estimate of this vertex; S5. Select vertex: Select the vertex with the smallest current minimum weight estimate value from the priority queue as the new current vertex; S6. Repeat steps S2 and S3: For the new current vertex, repeat the processes of updating the total path weight and selecting vertices in steps S2 and S3 until all vertices have been processed or the priority queue is empty; S7. Construct the relative shortest path: Use the information of the predecessor nodes to backtrack the relative shortest path from the evacuation starting point to the s safe point ; The set of road segments for all route trips between the evacuation points and each safe point is stored according to the flow direction of the traffic flow. All the total routes are L, and the road segments going to the s safe point are , .

9. An intelligent collaborative perception and decision-making system for emergency response of the Tibet access road, characterized in that, The expression of the multi-objective optimization model is: ； The specific constraint conditions are as follows: ； Among them, is the traffic flow prediction information of the road section, which will be continuously updated as the prediction information is updated, so as to achieve dynamic evacuation and scheduling; T is the total evacuation time, representing the overall time for evacuation vehicles to reach each safe point from the starting point. The objective function aims to minimize this parameter; D is the vehicle travel distance, representing the total distance or section length of the vehicle travel path, which together with the travel time affects the evacuation efficiency; V is the number of vehicles, representing the total number of vehicles to be evacuated or the number of vehicles allocated to each safe point s; α and β are weight coefficients used to balance the priorities between multiple objectives. α is used to weight the objective of minimizing time, and β is used to weight the objective of traffic capacity reliability; ns is the number of safe points, representing the total number of safe points in the system and used as an index in the constraint conditions; Ks is the capacity of safe point s, representing the maximum number of vehicles that safe point s can accommodate; fs is the traffic capacity of section s, representing the number of vehicles that can pass through section s per unit time; c is the environmental adjustment coefficient used to dynamically adjust the relationship between the traffic capacity of the section and the number of allocated vehicles, reflecting the attenuation effect of environmental factors on the traffic capacity; In the above formula, the objective function of formula (1) aims to minimize the overall time for evacuation vehicles to reach each safe point; The objective function of formula (2) ensures the evacuation speed and avoids secondary accidents caused by traffic congestion on evacuation roads through a traffic capacity reliability function of the section; In formula (3), ensure that the traffic capacity of the section leading to safe point s is greater than or equal to the number of allocated vehicles, and c is the environmental adjustment coefficient; In formula (4), all parameters are greater than 0; In formula (5), ensure that the number of vehicles allocated to each safe point is equal to the number of vehicles to be evacuated and dispatched; In formula (6), ensure that the number of vehicles at each safe point does not exceed the maximum capacity; In formula (7), ensure the consistency of the vehicle allocation path, which is consistent with the shortest path.

10. The intelligent collaborative perception and decision-making system for emergency response of the Tibet-inbound channel according to claim 1, characterized in that, The data of different departments and levels include accident disaster information, emergency support information, road network operation status information, and emergency rescue material information; The retrieval includes fast retrieval by keywords, complex condition retrieval, and association query.

Images