Expressway emergency response method and system based on control unit

Abstract

The application provides a kind of expressway emergency response method and system based on management and control unit, by obtaining the event logic knowledge network of expressway emergency event and the driving event data in the control unit monitoring flow, the knowledge vector data of each control unit driving event in multiple emergency response stages with logical time sequence order can be systematically analyzed and processed. By fusing the initial knowledge vector data of each emergency response stage and the prior transmission knowledge vector data, more accurate and comprehensive transmission knowledge vector data is generated, and the corresponding response scheme of each emergency response stage is determined. This method not only improves the accuracy and efficiency of emergency response, but also optimizes the allocation and scheduling of emergency resources by determining the priority information of candidate control unit driving events. Finally, according to the priority information of each control unit driving event, the disposal event corresponding to the expressway emergency event can be quickly and accurately determined.

Description

Technical Field

[0001] This invention relates to the field of computer technology, and more specifically, to a highway emergency response method and system based on a control unit. Background Technology

[0002] Emergency response is a crucial task in highway operation and management. As a vital transportation artery, the safety and smooth flow of highways directly impact public travel safety and economic benefits. However, various emergencies frequently occur on highways, such as traffic accidents, vehicle breakdowns, and severe weather. These events pose serious threats to the normal operation of highways, necessitating timely and effective emergency response.

[0003] Traditional emergency response methods often rely on human experience and rule-based judgment, which may be effective for simple, routine emergencies. However, with the continuous increase in highway traffic volume and the increasing complexity of emergencies, traditional methods are no longer sufficient to meet the demands for efficient and accurate emergency response. On the one hand, human experience and rule-based judgment are subjective and limited, making it difficult to comprehensively and accurately grasp the development trend and key points of emergency response. On the other hand, traditional methods lack real-time tracking and intelligent analysis of the dynamic changes in emergency events, making it difficult to adjust and optimize response plans in a timely manner during the emergency response process. Summary of the Invention

[0004] In view of the aforementioned problems, and in conjunction with the first aspect of the present invention, embodiments of the present invention provide a highway emergency response method based on a control unit, the method comprising:

[0005] Acquire the event logic knowledge network of highway emergency events and the driving event data corresponding to each driving event in the monitoring flow of each control unit;

[0006] Based on the event logic knowledge network and the driving event data corresponding to each control unit driving event, the initial knowledge vector data of each control unit driving event in multiple emergency response stages are obtained, and the multiple emergency response stages have a logical temporal order.

[0007] For each control unit driving event, the initial knowledge vector data of the candidate control unit driving event at each emergency response stage and the corresponding prior transmitted knowledge vector data are interactively fused to generate the transmitted knowledge vector data corresponding to each emergency response stage. The prior transmitted knowledge vector data are the transmitted knowledge vector data corresponding to the emergency response stages that have a logical temporal relationship.

[0008] Based on the knowledge vector data transmitted for each emergency response stage, the response plan for each emergency response stage is determined, and based on the response plan for each emergency response stage, the priority information of the candidate control unit driving events is determined.

[0009] Based on the priority information corresponding to the driving events of each control unit in the monitoring stream of the control unit, the handling events corresponding to the highway emergency events are determined.

[0010] In another aspect, embodiments of the present invention also provide a highway emergency response system based on a control unit, including a processor and a machine-readable storage medium connected to the processor. The machine-readable storage medium is used to store programs, instructions, or code, and the processor is used to execute the programs, instructions, or code in the machine-readable storage medium to implement the above-described method.

[0011] Based on the above, this application embodiment, by acquiring the event logic knowledge network of highway emergency events and the driving event data in the monitoring stream of control units, can systematically analyze and process the knowledge vector data of each control unit's driving events across multiple emergency response stages with logical temporal order. By fusing the initial knowledge vector data and prior transmitted knowledge vector data of each emergency response stage, more accurate and comprehensive transmitted knowledge vector data is generated, thereby determining the response plan corresponding to each emergency response stage. This method not only improves the accuracy and efficiency of emergency response but also optimizes the allocation and scheduling of emergency resources by determining the priority information of candidate control unit driving events. Finally, based on the priority information of each control unit's driving events, the corresponding handling events for highway emergency events can be quickly and accurately determined, effectively improving the intelligence level and handling efficiency of highway emergency response. Attached Figure Description

[0012] Figure 1 This is a schematic diagram of the execution flow of the highway emergency response method based on a control unit provided in an embodiment of the present invention.

[0013] Figure 2 This is a schematic diagram of the hardware architecture of a highway emergency response system based on a control unit, provided in an embodiment of the present invention. Detailed Implementation

[0014] The present invention will now be described in detail with reference to the accompanying drawings. Figure 1 This is a flowchart illustrating a highway emergency response method based on a control unit according to an embodiment of the present invention. The following is a detailed description of the highway emergency response method based on a control unit.

[0015] Step S110: Obtain the driving event data corresponding to each driving event in the event logic knowledge network and control unit monitoring stream of the highway emergency event.

[0016] In this embodiment, within a large-scale highway traffic management system, a highway emergency event is assumed, such as a multi-vehicle rear-end collision on a section of highway. This event logic knowledge network contains various logical relationship information related to the accident. For example, the location of the accident and the position of nearby rescue resources, the severity of the accident and the possible rescue forces required, and the logical relationship between the time of the accident and the impact on subsequent traffic flow. If the accident occurs near a traffic hub during peak traffic hours, the event logic knowledge network will include information such as the hub's diversion channels and peak-hour traffic flow patterns.

[0017] For the driving events of each control unit in the control unit monitoring stream, the control unit can be understood as a monitoring unit for different sections or functional areas on the highway. Taking the accident site as an example, each control unit monitors driving events within its respective area. Driving event data contains a wealth of information. For instance, if a control unit monitors a car suddenly decelerating 2 kilometers ahead of the accident site, its driving event data might include the vehicle's speed change curve (specific numerical records of the gradual decrease from normal speed), the vehicle type (whether it's a passenger car or a large truck), and the driver's behavior (whether there was emergency braking or abnormal lane changes). Another control unit might monitor a vehicle attempting an illegal U-turn behind the accident; its driving event data would include the vehicle's location, the time of the U-turn, and the vehicle's license plate number. By collecting data on the overall situation of highway emergency events and the monitoring data from each control unit, the data acquisition in this step is completed.

[0018] Step S120: Based on the event logic knowledge network and the driving event data corresponding to each control unit driving event, obtain the initial knowledge vector data of each control unit driving event in multiple emergency response stages, wherein the multiple emergency response stages have a logical temporal order.

[0019] Taking the aforementioned multi-vehicle rear-end collision on the highway as an example, the emergency response phase can be divided into the initial information collection and assessment phase, the rescue force deployment phase, the traffic diversion phase, and the recovery phase after the accident cleanup.

[0020] For the initial information collection and assessment phase of an accident, focusing on a driving event monitored by a specific control unit, such as the vehicle that suddenly decelerated 2 kilometers ahead of the accident mentioned earlier, the following knowledge encoding representations are first applied: the event logical knowledge network (including the logical relationship between the accident and the surrounding environment, rescue resources, etc.), the driving event data corresponding to the vehicle (vehicle speed change curve, vehicle type, etc.), and emergency response condition information (such as the need to quickly determine whether this stage will trigger subsequent chain accidents). For example, the distance relationship between the accident and rescue resources in the event logical knowledge network is encoded into a specific vector form, the vehicle type is encoded into vector elements according to a preset classification, and the need for rapid judgment in the emergency response conditions is also encoded into corresponding vector representations, thereby generating various knowledge encoding vectors.

[0021] Next, these knowledge encoding vectors are integrated. For example, accident-related logical relationship vectors, vehicle driving data vectors, and emergency response condition vectors are combined according to certain rules to form integrated knowledge encoding data. Then, each logical learning module performs logical reconstruction on this integrated knowledge encoding data. Assuming there are three logical learning modules, the first module focuses on handling logical relationships related to the accident location. Based on the accident location information in the integrated knowledge encoding data (such as the accident being located at a specific kilometer mark on a highway), it reconstructs logical relationships related to that location, such as relationships with nearby service areas and toll stations. The second module handles vehicle type-related logic, reconstructing the impact logic of different vehicle types in the accident. The third module handles emergency response condition-related logic, reconstructing the logic of how to quickly extract key information from the data based on the need for rapid judgment. This generates logical reconstruction data corresponding to each logical learning module.

[0022] Next, the weighting coefficients corresponding to each logical learning module are obtained. The importance space array is then spatially transformed based on the integrated knowledge encoding data. For example, the transformation of accident location information in the integrated knowledge encoding data within the importance space array might generate target transformation data related to location importance. If the accident occurred on a critical road segment prone to traffic congestion, this target transformation data might be larger. Stimulating this target transformation data (e.g., through a specific activation function) generates the weighting coefficients corresponding to each logical learning module.

[0023] Based on the target logic reconstruction data and corresponding target weighting coefficients for each target logic learning module corresponding to each emergency response stage, the fused knowledge vector data for each emergency response stage is determined. For the initial information collection and assessment stage, the target logic reconstruction data and corresponding target weighting coefficients for each target logic learning module are weighted. For example, the logic reconstruction data of the first logic learning module regarding the accident location is multiplied by its weighting coefficient, the logic reconstruction data of the second logic learning module regarding vehicle type is multiplied by its weighting coefficient, and so on, to obtain the weighted calculation result for each target logic learning module. These weighted calculation results are then fused, for example, by adding them according to certain weights or using a more complex fusion algorithm, to generate the fused knowledge vector data corresponding to that emergency response stage. Finally, self-attention processing is applied to the fused knowledge vector data corresponding to each emergency response stage. For example, in the initial stage of an accident, self-attention processing adjusts the weights of different elements in the fused knowledge vector data according to their importance (e.g., the accident location element may be more important), generating the initial knowledge vector data for the control unit's driving events in this emergency response stage. The same process will be carried out in multiple emergency response phases, such as the deployment of rescue forces, traffic management, and post-accident recovery, to obtain initial knowledge vector data of driving events in each control unit at each emergency response phase.

[0024] Step S130: For each control unit driving event, the initial knowledge vector data and the corresponding prior transmitted knowledge vector data of the candidate control unit driving event at each emergency response stage are interactively fused to generate the transmitted knowledge vector data corresponding to each emergency response stage. The prior transmitted knowledge vector data are the transmitted knowledge vector data corresponding to the emergency response stages that have a logical temporal relationship.

[0025] Taking the rescue force deployment phase as an example, for a driving event within a specific control unit, such as the previously mentioned incident where a vehicle attempted to illegally make a U-turn behind an accident, the initial knowledge vector data at this stage includes vector representations of the vehicle's illegal U-turn location, time, and related vehicle information, processed through previous steps. The corresponding prior transferred knowledge vector data might be transferred knowledge vector data related to the overall accident situation generated during the initial accident information collection and assessment phase.

[0026] First, a knowledge fusion coefficient is generated based on the initial knowledge vector data of the candidate control unit's driving event during the rescue force deployment phase and the corresponding prior transmitted knowledge vector data. This knowledge fusion coefficient may be generated based on correlation analysis between the two. For example, if the location of the vehicle's illegal U-turn in the initial knowledge vector data is close to the core accident area in the prior transmitted knowledge vector data, this correlation may result in a larger knowledge fusion coefficient.

[0027] Then, based on the initial knowledge vector data of the candidate control unit's driving event during the rescue force deployment phase, the corresponding prior transmitted knowledge vector data, and the knowledge fusion coefficient, reference knowledge vector data is generated. For example, the initial knowledge vector data and the prior transmitted knowledge vector data are combined according to the knowledge fusion coefficient using a specific calculation method (such as weighted summation or vector multiplication) to obtain the reference knowledge vector data.

[0028] Finally, based on the knowledge fusion coefficient, the initial knowledge vector data of the candidate control unit driving event during the rescue force allocation phase, and the reference knowledge vector data, transferable knowledge vector data corresponding to the rescue force allocation phase is generated. This transferable knowledge vector data may contain more information about how to allocate rescue forces to handle illegally turning vehicles and the coordination relationship with the overall accident rescue. The same process is carried out in other emergency response phases such as traffic diversion and post-accident recovery. For each control unit driving event, its initial knowledge vector data and corresponding prior transferable knowledge vector data in each emergency response phase are interactively fused in the above manner to generate transferable knowledge vector data corresponding to each emergency response phase.

[0029] Step S140: Based on the knowledge vector data corresponding to each emergency response stage, determine the response plan corresponding to each emergency response stage, and based on the response plan corresponding to each emergency response stage, determine the priority information of the candidate control unit driving events.

[0030] During the rescue force deployment phase, the response plan is determined based on the transferred knowledge vector data corresponding to this phase. This transferred knowledge vector data includes information such as the severity of the accident, the distribution of rescue resources around the accident area, traffic conditions at the accident site, and relevant information on driving events from various control units. If the transferred knowledge vector data indicates a severe accident and that rescue resources are far away, and a driving event from a certain control unit involves a large rescue vehicle approaching the accident site, the response plan might be to prioritize guiding that rescue vehicle to the accident site quickly, while simultaneously coordinating surrounding traffic control to ensure unobstructed rescue access.

[0031] During the traffic management phase, if the transmitted knowledge vector data indicates severe traffic congestion behind the accident site and multiple control unit events involving vehicle flows in different directions, the response plan might be to triage and divert vehicles behind the accident site, prioritizing the evacuation of passenger cars through emergency lanes, and temporarily controlling large trucks pending further arrangements.

[0032] A global analysis of the response plans corresponding to each emergency response phase is conducted to determine the priority information of candidate control unit (CMU) driving events. For example, during the entire emergency response process, CMU driving events directly related to rescue (such as the movement of rescue vehicles) may be given higher priority because they play a crucial role in resolving the core issues of the accident. Conversely, some CMU driving events with less impact on traffic (such as events involving vehicles traveling normally away from the accident area) may be given lower priority. By comprehensively considering the response plans for different emergency response phases, the priority information of each candidate CMU driving event is accurately determined.

[0033] Step S150: Based on the priority information corresponding to the driving events of each control unit in the control unit monitoring stream, determine the handling event corresponding to the highway emergency event.

[0034] Continuing with the example of the multi-vehicle rear-end collision on the highway, the monitoring stream of the control unit contains multiple control unit driving events, such as rescue vehicle driving events, normal driving events of vehicles around the accident, and illegal driving events. The appropriate action is determined based on the priority information corresponding to each of these control unit driving events.

[0035] If the incident involving rescue vehicles has the highest priority, then the primary task in handling highway emergency incidents is to ensure that rescue vehicles can reach the accident scene smoothly to carry out rescue work. This may include opening a dedicated lane for rescue vehicles and coordinating traffic lights and traffic control measures along the route. For incidents involving vehicles violating traffic rules, if their priority is high (e.g., violations severely impacting rescue or traffic flow), the handling may involve promptly stopping the violation, penalizing the offending driver, and guiding the violating vehicle to a safe area to prevent further interference with the accident scene and rescue work. For incidents involving vehicles traveling normally around the accident, since their priority is relatively low, the handling may involve minimizing the impact on their normal driving while ensuring rescue and traffic flow, such as using appropriate traffic guidance signs and broadcast announcements to guide normal vehicles to detour in an orderly manner or slow down.

[0036] Based on the activation of the emergency response interface, the response events corresponding to the highway emergency events are presented in priority order. For example, on the emergency response center's operation interface, when the operator activates the emergency response interface, the response events corresponding to the highest priority control unit driving events are displayed first, such as guidance and support measures for rescue vehicles. Then, the response events corresponding to other priority control unit driving events are displayed in sequence, so that the operator can carry out emergency response work in an orderly manner according to priority.

[0037] Based on the above steps, this embodiment of the application, by acquiring the event logic knowledge network of highway emergency events and the driving event data in the monitoring stream of the control unit, can systematically analyze and process the knowledge vector data of each control unit's driving event across multiple emergency response stages with logical temporal order. By fusing the initial knowledge vector data and prior transmitted knowledge vector data of each emergency response stage, more accurate and comprehensive transmitted knowledge vector data is generated, thereby determining the response plan corresponding to each emergency response stage. This method not only improves the accuracy and efficiency of emergency response but also optimizes the allocation and scheduling of emergency resources by determining the priority information of candidate control unit driving events. Finally, based on the priority information of each control unit's driving events, the corresponding handling events for highway emergency events can be quickly and accurately determined, effectively improving the intelligence level and handling efficiency of highway emergency response.

[0038] In one possible implementation, step S120 includes:

[0039] Step S121: For each driving event of a control unit, the event logic knowledge network, the driving event data corresponding to the candidate control unit driving event, and the emergency response condition information are represented by knowledge encoding to generate various knowledge encoding vectors.

[0040] Step S122: Integrate the various knowledge encoding vectors to generate integrated knowledge encoding data.

[0041] Step S123: The integrated knowledge encoding data is logically reconstructed by each logical learning module to generate logical reconstruction data corresponding to each logical learning module.

[0042] Step S124: Obtain the weighting coefficients corresponding to each logical learning module.

[0043] Step S125: Based on the target logic reconstruction data and target weighting coefficients corresponding to each target logic learning module for each emergency response stage, determine the fusion knowledge vector data corresponding to each emergency response stage.

[0044] Step S126: Perform self-attention processing on the fused knowledge vector data corresponding to each emergency response stage to generate initial knowledge vector data of the candidate control unit driving events at each emergency response stage.

[0045] In this embodiment, taking the previously mentioned multi-vehicle rear-end collision on a highway as an example, for each control unit driving event, such as the driving event of a vehicle suddenly decelerating 2 kilometers ahead of the accident, the first step is to perform knowledge encoding representation on the event logic knowledge network, the driving event data corresponding to the candidate control unit driving events, and the emergency response condition information to generate various knowledge encoding vectors. The event logic knowledge network contains the logical relationships between the accident and various surrounding factors, such as the location relationship between the accident site and rescue facilities, traffic hubs, etc., and these relationships are converted into vector form according to specific encoding rules. For example, if the accident site is 5 kilometers away from the nearest rescue station, this distance relationship may be encoded as a vector element with a specific value. Regarding the driving event data, the speed change curve in the event of a vehicle suddenly decelerating can be encoded into vector elements at certain time intervals. If the vehicle type is a small passenger car, it is converted into vector elements according to the pre-set small passenger car encoding rules. The emergency response condition information may be used to quickly determine whether the accident will trigger subsequent chain reactions in the early stages of the accident. This condition is also encoded according to the corresponding rules, ultimately forming various knowledge encoding vectors.

[0046] Next, the event logic knowledge network encoding vector, the driving event data encoding vector, and the emergency response condition encoding vector are combined according to a predetermined combination method. For example, the accident location-related elements in the event logic knowledge network encoding vector are first associated and combined with the vehicle location elements in the driving event data encoding vector, and then the judgment requirement elements in the emergency response condition encoding vector are integrated with other relevant elements to form integrated knowledge encoding data.

[0047] Then, each logic learning module performs logical reconstruction on the integrated knowledge-encoded data, generating corresponding logical reconstruction data for each module. Assuming there are three logic learning modules, the first module focuses on reconstructing the logical relationship between the accident site and its surrounding environment. It obtains location information about the accident site and surrounding areas, such as service areas and toll stations, from the integrated knowledge-encoded data, reconstructing a more comprehensive logical relationship related to the accident site. For example, if the accident site is between two toll stations and close to a service area exit, the reconstructed logical relationship will detail the potential impact of this locational relationship on accident handling, such as which direction is more convenient for rescue vehicles to enter. The second logic learning module focuses on vehicle-related logic, such as reconstructing the vehicle's driving logic in an accident scenario based on information like sudden deceleration and vehicle type, such as whether the sudden deceleration of a small passenger car might be due to obstructed forward visibility. The third logic learning module addresses emergency response condition logic. Based on the emergency response judgment requirements in the integrated knowledge-encoded data, it reconstructs the logic of how to efficiently obtain the necessary information from the data, such as which data elements are most critical for determining whether a chain reaction of accidents will occur, thus generating corresponding logical reconstruction data for each logic learning module.

[0048] Next, the weighting coefficients corresponding to each logical learning module are obtained. Based on the integrated knowledge encoding data, the importance spatial array is spatially transformed to generate target transformation data matching the number of logical learning modules. Different elements in the integrated knowledge encoding data have different levels of importance. For example, accident location information may have a high importance weight in the overall emergency response. When spatially transforming the importance spatial array based on the integrated knowledge encoding data, if the accident location is near a transportation hub, this important location information may generate a larger target transformation data after spatial transformation. Then, the target transformation data is stimulated to generate the weighting coefficients corresponding to each logical learning module. This stimulation process is carried out according to a pre-set stimulation function; different target transformation data are calculated using the stimulation function to obtain the weighting coefficients of the corresponding logical learning modules.

[0049] Based on the target logic reconstruction data and corresponding target weighting coefficients for each target logic learning module at each emergency response stage, the fused knowledge vector data for each emergency response stage is determined. Taking the initial emergency response stage as an example, for each target logic learning module, its target logic reconstruction data and corresponding target weighting coefficients are weighted and calculated. For example, the first logic learning module's logic reconstruction data regarding the accident location is multiplied by its weighting coefficient, the second logic learning module's logic reconstruction data regarding vehicle movement is multiplied by its weighting coefficient, and so on, to obtain the weighted calculation result for each target logic learning module. These weighted calculation results are then fused. This fusion process is not a simple addition, but rather a specific fusion algorithm is used based on the characteristics of each emergency response stage. For example, in the initial stage of an accident, more emphasis may be placed on the logical relationship between the accident location and vehicle movement, so the fusion algorithm will give these two aspects higher weights in the weighted calculation results. After fusion processing, the fused knowledge vector data corresponding to this emergency response stage is generated.

[0050] Finally, self-attention processing is applied to the fused knowledge vector data corresponding to each emergency response stage to generate initial knowledge vector data for candidate control unit driving events at each emergency response stage. In the initial emergency response stage, the fused knowledge vector data contains information elements such as accident location, vehicle movement, and emergency response logic. Self-attention processing automatically adjusts the weights of these elements based on their importance in the emergency response at that stage. For example, the accident location element is crucial in determining the scope of the accident's impact and planning rescue routes, so self-attention processing increases its weight, while the vehicle type element, which is relatively less important at this stage, may have its weight reduced. After this adjustment, initial knowledge vector data for candidate control unit driving events in the initial emergency response stage is generated. The same process is performed in other emergency response stages, such as the rescue force deployment stage, traffic control stage, and post-accident recovery stage, to obtain initial knowledge vector data for each control unit driving event at each emergency response stage.

[0051] In one possible implementation, step S124 includes:

[0052] Step S1241: Based on the integrated knowledge encoding data, perform spatial transformation on the importance space array to generate target transformation data that matches the number of logical learning modules.

[0053] Step S1242: The target transformation data is stimulated to generate weighting coefficients corresponding to each logical learning module.

[0054] In this embodiment, in the previously described scenario of a multi-vehicle pile-up on a highway, the integrated knowledge-encoded data contains various information elements, such as the location of the accident, the driving data of the vehicles involved, and emergency response conditions. The importance spatial array is a pre-defined structure used to measure the importance of different information elements. For example, in a multi-vehicle pile-up scenario, information such as the precise location of the accident and its severity has a high importance weight in the initial stage of emergency response, while some minor driving information (such as minor speed fluctuations of normally moving vehicles) has relatively low importance. When the importance spatial array is spatially transformed based on the integrated knowledge-encoded data, taking accident location information as an example, if the accident occurs on a critical section of the highway, such as a section with high traffic volume connecting multiple important areas, then the accident location-related elements in the integrated knowledge-encoded data will generate a larger target transformation data when spatially transformed with the importance spatial array. Suppose there are three logical learning modules: accident location logical learning module, vehicle driving logical learning module, and emergency response condition logical learning module. For each logical learning module, spatial transformation is performed based on the corresponding elements and importance spatial array in the integrated knowledge encoding data, thereby generating target transformation data that matches the number of logical learning modules.

[0055] The target transformation data is stimulated to generate weighted coefficients for each logical learning module. This stimulation operation is based on a specific activation function. Different target transformation data will yield different results after calculation by the activation function, and these results are the weighted coefficients for each logical learning module. For example, for the accident location logical learning module, if its target transformation data reflects the high importance of the accident location (e.g., the accident occurred on a bottleneck section of a highway), a large weighted coefficient will be obtained after calculation by the activation function. This indicates that the output of the accident location logical learning module will be given a higher weight in subsequent logical processing. Similarly, for the vehicle driving logical learning module and the emergency response condition logical learning module, their respective target transformation data will obtain corresponding weighted coefficients after calculation by the activation function. These weighted coefficients will affect their role in the subsequent determination of fused knowledge vector data.

[0056] In one possible implementation, step S125 includes:

[0057] Step S1251: For each emergency response stage, the target logic reconstruction data and corresponding target weighting coefficients of each target logic learning module corresponding to the candidate emergency response stage are weighted and calculated to generate the weighted calculation result of each target logic learning module corresponding to the candidate emergency response stage.

[0058] Step S1252: The weighted calculation results of each target logic learning module corresponding to the candidate emergency response stage are fused to generate fused knowledge vector data corresponding to the candidate emergency response stage.

[0059] In detail, for each emergency response phase, such as the initial emergency response phase of an accident, the target logic reconstruction data and corresponding target weighting coefficients of each target logic learning module corresponding to the candidate emergency response phase are weighted and calculated to generate the weighted calculation result for each target logic learning module corresponding to the candidate emergency response phase. In the initial stage of an accident, for the accident location logic learning module, its target logic reconstruction data may include detailed logical relationships between the accident site and surrounding rescue resources and traffic facilities. Assuming its target weighting coefficient is a high value (based on the previous calculation), each element in the target logic reconstruction data is multiplied by the target weighting coefficient to obtain the weighted calculation result for that target logic learning module. For the vehicle driving logic learning module, its target logic reconstruction data may include logical relationships such as the driving trajectory and speed changes of accident-related vehicles; similarly, it is weighted and calculated with its own target weighting coefficient to obtain the weighted calculation result. The emergency response condition logic learning module is also operated in the same way to obtain its weighted calculation result.

[0060] The weighted calculation results of each target logic learning module corresponding to a candidate emergency response stage are fused to generate fused knowledge vector data for that stage. This fusion process is not a simple addition or averaging operation, but rather employs a specific fusion algorithm tailored to the characteristics and needs of each emergency response stage. In the initial emergency response stage, where the determination of accident location and emergency response conditions is relatively more critical, the fusion algorithm assigns higher weights to the weighted calculation results of the accident location logic learning module and the emergency response condition logic learning module. For example, the fusion algorithm might be a weighted summation method, where the weighted calculation result of the accident location logic learning module is set to 0.4, the weighted calculation result of the emergency response condition logic learning module is set to 0.3, and the weighted calculation result of the vehicle driving logic learning module is set to 0.3. Following this weight allocation, the weighted calculation results of the three target logic learning modules are fused to obtain the fused knowledge vector data corresponding to the initial emergency response stage. This fused knowledge vector data contains a knowledge representation that comprehensively considers information from multiple aspects such as accident location, vehicle driving, and emergency response conditions, fused according to specific weights, and can more accurately reflect the overall situation in the initial emergency response stage. The same process will be carried out in other emergency response phases, such as the deployment of rescue forces, traffic management, and post-accident recovery. For each emergency response phase, the corresponding fused knowledge vector data will be determined through such weighted calculation and fusion processing.

[0061] In one possible implementation, step S130 includes:

[0062] For each emergency response phase, a knowledge fusion coefficient is generated based on the initial knowledge vector data of the candidate control unit driving event in the candidate emergency response phase and the corresponding prior transmitted knowledge vector data.

[0063] Reference knowledge vector data is generated based on the initial knowledge vector data of the candidate control unit driving event in the candidate emergency response phase, the corresponding prior transmitted knowledge vector data, and the knowledge fusion coefficient.

[0064] Based on the knowledge fusion coefficient, the initial knowledge vector data of the candidate control unit driving event in the candidate emergency response stage, and the reference knowledge vector data, the transferable knowledge vector data corresponding to the candidate emergency response stage is generated.

[0065] In this embodiment, in the scenario of a multi-vehicle rear-end collision on a highway, operations are performed on driving events for each control unit during the rescue force deployment phase. For example, for a driving event monitored by a certain control unit where a vehicle attempts to illegally make a U-turn behind the accident, there is corresponding initial knowledge vector data during the rescue force deployment phase. This initial knowledge vector data includes vector representations of information such as the location of the illegally making U-turn vehicle (marked with the precise kilometer number of the highway), the type of vehicle (e.g., a small passenger car), and the time of the illegal U-turn, processed in previous steps. The corresponding prior transferred knowledge vector data is transferred knowledge vector data related to the overall situation of the accident, generated during the initial information collection and assessment phase of the accident. This data may include information such as the severity of the accident (e.g., the number of vehicles involved, casualties, etc.) and the traffic congestion range at the accident site (represented by the number of kilometers extending forward and backward from the accident point).

[0066] First, a knowledge fusion coefficient is generated based on the initial knowledge vector data of the candidate control unit's driving event during the rescue force deployment phase and the corresponding prior transmitted knowledge vector data. This process is achieved by analyzing various relationships between the two. For example, considering the relationship between the location of the illegally making U-turn vehicle in the initial knowledge vector data and the traffic congestion range of the accident scene in the prior transmitted knowledge vector data, if the illegally making U-turn vehicle is located in the core area of ​​the traffic congestion range or is about to enter this core area, it indicates a high degree of correlation between the two. This correlation will be reflected in the knowledge fusion coefficient, potentially resulting in a larger coefficient. Simultaneously, the vehicle type in the initial knowledge vector data also affects the knowledge fusion coefficient. If a large vehicle illegally makes a U-turn, its impact on rescue force deployment may be greater than that of a small vehicle, because large vehicles may occupy more road space. Therefore, this factor will be considered when calculating the knowledge fusion coefficient, and the coefficient will be appropriately increased. Through comprehensive analysis and calculation of these relationships, a specific knowledge fusion coefficient is ultimately generated.

[0067] Next, based on the initial knowledge vector data of the candidate control unit's driving event during the rescue force deployment phase, the corresponding previously transmitted knowledge vector data, and the knowledge fusion coefficient, reference knowledge vector data is generated. Assume the initial knowledge vector data is represented as vector A, the previously transmitted knowledge vector data as vector B, and the knowledge fusion coefficient is k. The process of generating reference knowledge vector data can employ a specific calculation method, for example, reference knowledge vector data C = k*A + (1-k)*B. In this calculation process, the knowledge fusion coefficient k determines the weight allocation between the initial knowledge vector data A and the previously transmitted knowledge vector data B when generating reference knowledge vector data C. If k is large, it indicates that the initial knowledge vector data A plays a more important role in generating reference knowledge vector data C; conversely, if k is small, the previously transmitted knowledge vector data B has a more significant effect. Taking the previously mentioned illegal U-turn vehicle as an example, if the location of the illegal U-turn vehicle is closely related to the traffic congestion area at the accident scene (i.e., k is large), then when generating the reference knowledge vector data, the location information and vehicle type of the illegal U-turn vehicle in the initial knowledge vector data will be more prominently reflected in the reference knowledge vector data. At the same time, the severity of the accident and the scope of traffic congestion in the prior knowledge vector data will also participate in the composition of the reference knowledge vector data according to their respective weights.

[0068] Finally, based on the knowledge fusion coefficient, the initial knowledge vector data of the candidate control unit driving event in the candidate emergency response phase, and the reference knowledge vector data, the transferable knowledge vector data corresponding to the candidate emergency response phase is generated. This process may involve more complex calculations and logical processing. For example, the transferable knowledge vector data D can be generated as follows: D = f(k, A, C), where f is a specific function designed based on the needs of the emergency response phase and the inherent logical relationship between the data. In the rescue force allocation phase, this function may place greater emphasis on the allocation of rescue resources and traffic control measures. Taking illegally turning vehicles as an example, the function f will generate transferable knowledge vector data D containing information such as how to allocate rescue forces to handle illegally turning vehicles and how to coordinate with the overall accident rescue relationship, based on the knowledge fusion coefficient k, the relevant information of illegally turning vehicles in the initial knowledge vector data A, and the comprehensive information in the reference knowledge vector data C. For example, if the reference knowledge vector data C indicates that there is severe traffic congestion and limited rescue resources at the accident scene, while the initial knowledge vector data A shows that vehicles making illegal U-turns are further obstructing traffic, then the transmitted knowledge vector data D may contain information such as prioritizing the dispatch of traffic control personnel to the location of the illegally making U-turns to guide traffic and adjusting the route of rescue forces to avoid the obstructing area.

[0069] The same process is conducted in other emergency response phases, such as traffic management and post-accident recovery. In the traffic management phase, the initial knowledge vector data may contain information such as vehicle direction and speed, while the prior knowledge vector data includes information such as the progress of accident scene cleanup and current traffic flow distribution. Following the steps described above, knowledge fusion coefficients, reference knowledge vector data, and the final transferred knowledge vector data are generated. In the post-accident recovery phase, the initial knowledge vector data may involve the driving status of remaining vehicles, while the prior knowledge vector data includes information such as the status of road infrastructure repairs. The same steps are followed to generate the corresponding transferred knowledge vector data, thus providing accurate and comprehensive transferred knowledge vector data for each emergency response phase, facilitating subsequent determination of the appropriate response plan.

[0070] In one possible implementation, step S140 may include:

[0071] A global analysis is performed on the response plans corresponding to each emergency response stage to generate priority information for candidate control unit driving events.

[0072] In this embodiment, there are multiple emergency response stages in the emergency handling of multi-vehicle rear-end collisions on highways, such as the initial information collection and assessment stage, the rescue force deployment stage, the traffic diversion stage, and the recovery stage after the accident cleanup. Each stage has a corresponding response plan.

[0073] In the initial information gathering and assessment phase of an accident, the response plan primarily revolves around rapidly obtaining detailed information about the accident. This includes determining the precise location of the accident (down to the exact kilometer mark and lane position on the highway), the number of vehicles involved, vehicle types (distinguishing between passenger cars, trucks, hazardous materials transport vehicles, etc.), and the extent of injuries or fatalities. Simultaneously, a preliminary assessment of the traffic conditions at the accident scene is necessary, including the impact range of the accident on surrounding traffic flow (the length of congestion extending forward and backward from the accident point) and the severity of traffic congestion (whether it causes complete road blockage or merely slowed traffic). A global analysis of the response plan at this stage determines the priority of driving events for candidate control units. For example, if a control unit driving event involves an emergency patrol vehicle equipped with monitoring equipment approaching the accident scene, this vehicle's driving event has a high priority because its monitoring equipment can provide crucial data for information gathering and assessment, helping to gain a more comprehensive and accurate understanding of the accident situation. Conversely, vehicles traveling normally away from the accident scene have a lower priority because their contribution to accident information gathering and assessment at this stage is relatively small.

[0074] In the rescue force deployment phase, the response plan focuses on allocating rescue resources, such as route planning and time scheduling for fire trucks, medical rescue vehicles, and rescue engineering vehicles to reach the accident site. It also includes coordinating the cooperation between various rescue forces to ensure efficient and orderly rescue operations. During this phase, when conducting a global analysis to determine priorities, the driving events of control units directly related to rescue forces have the highest priority. For example, if a large rescue engineering vehicle carrying specialized rescue equipment is heading to the accident site, its driving event has a high priority because its timely arrival is crucial for the rescue operation. Vehicles traveling normally around the accident site but not involved in the rescue have a lower priority; these vehicles may require traffic control or guidance based on rescue needs to ensure unobstructed rescue routes.

[0075] During the traffic management phase, the response plan primarily targets traffic flow at and around the accident site to prevent further congestion. This includes implementing temporary traffic control measures, such as setting up diversion points and planning detour routes for vehicles. When conducting a global analysis of the response plan to determine priorities, driving events affecting key traffic management nodes have higher priority. For example, vehicles violating traffic rules or failing to follow guidance near diversion points have a high priority because they could disrupt the entire traffic management plan. Vehicles driving normally away from key traffic management areas have a relatively lower priority.

[0076] During the recovery phase following an accident, the response plan focuses on inspecting the repair status of road facilities and fully restoring traffic order. Events involving engineering vehicles inspecting road facilities have a high priority because they directly affect the safe return of roads to normal use. Events involving ordinary vehicles driving normally during this phase have a lower priority if they do not affect road facility inspections or traffic restoration. By conducting this global analysis of the response plans for each emergency response phase, the priority information for driving events in candidate control units can be accurately generated.

[0077] In one possible implementation, step S120 may further include:

[0078] For each control unit driving event, based on the event logic knowledge network and the driving event data corresponding to the candidate control unit driving event, the initial knowledge vector data of the candidate control unit driving event at multiple emergency response stages are obtained through the initial logic learning module in the emergency response analysis model.

[0079] Step S130 may include:

[0080] For each emergency response phase, the knowledge transfer and fusion module corresponding to the candidate emergency response phase in the emergency response analysis model is used to interactively fuse the initial knowledge vector data of the candidate control unit driving event in the candidate emergency response phase with the corresponding prior transferred knowledge vector data to generate the transferred knowledge vector data corresponding to the candidate emergency response phase.

[0081] For each control unit driving event, based on the event logic knowledge network and the driving event data corresponding to the candidate control unit driving events, the initial knowledge vector data of the candidate control unit driving events at multiple emergency response stages are obtained through the initial logic learning module in the emergency response analysis model.

[0082] Taking a driving event monitored by a control unit near the accident scene as an example, such as a vehicle driving normally in the adjacent lane to the accident site, the event logic knowledge network includes the overall layout information of the highway (number of lanes, entrance and exit locations, etc.) and the relationship between the accident and the surrounding environment (such as the distance and location relationship between the accident point and nearby service areas and toll stations). The driving event data corresponding to the driving event of the candidate control unit includes information such as the vehicle's speed, direction of travel, and vehicle type.

[0083] The initial logic learning module in the emergency response analysis model begins processing this information. During the initial information gathering and assessment phase of an accident, the initial logic learning module extracts information related to the accident location from the event logic knowledge network, such as the relationship between the accident point and adjacent lanes, and obtains vehicle location information (relative distance to the accident point) and speed from driving event data. Then, according to specific encoding rules and algorithms, this information is transformed into initial knowledge vector data. For example, the distance relationship between the accident point and adjacent lanes is encoded as a specific numerical element, and vehicle speed is transformed into an element in the vector according to a certain ratio.

[0084] During the rescue force deployment phase, the initial logic learning module retrieves logical information related to rescue resource allocation from the event logic knowledge network, such as information on areas around the accident site where rescue vehicles can park. It also obtains vehicle type information from driving event data (because different types of vehicles have different impacts on rescue access). Then, based on this information, it constructs the initial knowledge vector data for this phase. For example, if the vehicle is a large truck, it might be represented in the vector data with a specific encoding, indicating its potential obstruction of rescue access.

[0085] Similarly, during the traffic management phase and the recovery phase after accident clearance, the initial logic learning module will obtain the initial knowledge vector data of each control unit's driving events in these emergency response phases based on the event logic knowledge network and relevant information in the driving event data, according to different logical relationships and encoding methods.

[0086] For each emergency response phase, the knowledge transfer and fusion module corresponding to the candidate emergency response phase in the emergency response analysis model is used to interactively fuse the initial knowledge vector data of the candidate control unit driving event in the candidate emergency response phase with the corresponding prior transferred knowledge vector data to generate the transferred knowledge vector data corresponding to the candidate emergency response phase.

[0087] Taking the rescue force deployment phase as an example, for a driving event within a specific control unit, such as the vehicles traveling normally in adjacent lanes at the accident scene mentioned earlier, there is corresponding initial knowledge vector data at this stage. This data includes information such as vehicle type and relative position to the accident point. The corresponding prior transferred knowledge vector data is the transferred knowledge vector data related to the overall situation of the accident, generated during the initial information collection and assessment phase of the accident, such as the severity of the accident and the extent of traffic congestion at the accident scene.

[0088] The knowledge transfer and fusion module corresponding to the rescue force deployment phase in the emergency response analysis model begins operation. First, this module analyzes the initial knowledge vector data and the previously transferred knowledge vector data, generating a knowledge fusion coefficient based on the inherent logical relationship between them. For example, if the relative position of the vehicle to the accident site in the initial knowledge vector data falls within the traffic congestion area in the previously transferred knowledge vector data, and the vehicle type is a large vehicle, a larger knowledge fusion coefficient may be generated because large vehicles have a greater impact on rescue force deployment within congestion areas.

[0089] Next, based on the initial knowledge vector data of the candidate control unit's driving event during the rescue force deployment phase, the corresponding previously transmitted knowledge vector data, and this knowledge fusion coefficient, reference knowledge vector data is generated. Assuming the initial knowledge vector data is vector A, the previously transmitted knowledge vector data is vector B, and the knowledge fusion coefficient is k, the reference knowledge vector data C may be obtained through a specific calculation method, such as C = k*A + (1-k)*B. In this calculation process, the knowledge fusion coefficient k determines the weight allocation of vectors A and B when generating the reference knowledge vector data C.

[0090] Finally, based on the knowledge fusion coefficient, the initial knowledge vector data of the candidate control unit's driving events in the candidate emergency response phase, and the reference knowledge vector data, the transferable knowledge vector data corresponding to the candidate emergency response phase is generated. For example, a specific function f(k, A, C) is used to generate the transferable knowledge vector data. In the rescue force deployment phase, this function will generate transferable knowledge vector data containing information such as the assessment of the vehicle's impact on rescue force deployment and whether vehicle control or guidance is needed, based on the knowledge fusion coefficient k, vehicle-related information in the initial knowledge vector data A, and comprehensive information in the reference knowledge vector data C.

[0091] The same process will be carried out in other emergency response phases, such as traffic management and post-accident recovery. During traffic management, the knowledge transfer and fusion module will generate corresponding transferable knowledge vector data based on information such as vehicle direction and speed from the initial knowledge vector data and information such as the progress of accident scene cleanup and current traffic flow distribution from the previously transferred knowledge vector data, following the steps described above. During post-accident recovery, the knowledge transfer and fusion module will also generate transferable knowledge vector data for this phase based on information such as vehicle driving status from the initial knowledge vector data and road infrastructure repair status from the previously transferred knowledge vector data.

[0092] In one possible implementation, the step of using the knowledge transfer fusion module corresponding to the candidate emergency response stage in the emergency response analysis model to interactively fuse the initial knowledge vector data of the candidate control unit driving event at the candidate emergency response stage with the corresponding prior transferred knowledge vector data to generate the transferred knowledge vector data corresponding to the candidate emergency response stage includes:

[0093] Step S131: Normalize the initial knowledge vector data and the prior transmitted knowledge vector data, and extract first feature information related to emergency response from the normalized initial knowledge vector data to generate an initial feature set. The first feature information includes event type features, event severity features, and event location features. Also, extract second feature information related to the history of emergency response from the normalized prior transmitted knowledge vector data to generate a prior feature set. The second feature information includes the handling methods of similar historical events and the impact of historical events on the current event.

[0094] In this embodiment, taking the rescue force deployment phase as an example, for a certain control unit driving event, it is assumed that its initial knowledge vector data includes information such as the vehicle's driving speed near the accident section, vehicle type (e.g., large truck), and distance from the accident scene. The prior transmitted knowledge vector data includes information such as the severity of the accident (e.g., number of vehicles involved, casualties), and the extent of traffic congestion after the accident. After normalization, the driving speed in the initial knowledge vector data is converted into a value within the range [0, 1] according to certain rules. The vehicle type can be normalized according to a pre-set encoding method, and the distance from the accident scene is also converted into a specific normalized value. The first feature information is extracted from the normalized initial knowledge vector data: the event type feature is a driving event of a normally driving vehicle near the accident; the event severity feature can infer the potential impact on rescue based on information such as vehicle type and driving speed; and the event location feature is the vehicle's position relative to the accident scene. For the prior knowledge vector data, the characteristics of how similar historical events were handled might be empirical data on the deployment of rescue forces under similar accident scales in the past, such as how many fire trucks and medical rescue vehicles were deployed first. The characteristics of the impact of historical events on the current event could be an assessment of the congestion development trend of the current accident based on the traffic congestion spread of past accidents. In this way, the initial feature set and the prior feature set are generated respectively.

[0095] Step S132: Perform matching analysis on the features in the initial feature set and the prior feature set to determine the feature pairs that are related to each other in the initial feature set and the prior feature set. For target feature pairs with a correlation degree greater than a set threshold, analyze the internal logical relationship of the target feature pairs, and construct a feature correlation matrix based on the analysis results of the internal logical relationship. The feature correlation matrix is ​​used to record the correlation relationship and correlation strength between features.

[0096] For example, during the rescue force deployment phase, the vehicle type (large truck) in the initial feature set may be correlated with the accident severity (involving multiple vehicles) in the prior feature set. If a large truck is near an accident and the accident is severe, it may significantly obstruct rescue access, and this correlation may be high, exceeding a set threshold. Similarly, the distance between vehicles in the initial feature set and the accident scene may be correlated with the impact of historical events on the current event (traffic congestion spread trend) in the prior feature set. If a vehicle is close to the accident scene, and historical experience suggests traffic congestion may spread in that direction, this is also a highly correlated feature pair. For these target feature pairs, their inherent logical relationships are analyzed. Taking vehicle type (large truck) and accident severity (involving multiple vehicles) as an example, the inherent logical relationship is that large trucks, due to their size, occupy more road space in accidents involving multiple vehicles, thus obstructing the passage of rescue vehicles. Based on the analysis results of these inherent logical relationships, a feature correlation matrix is ​​constructed, recording the correlation between features (such as the correlation between vehicle type and accident severity) and the correlation strength (quantified numerically based on the specific correlation degree).

[0097] Step S133: Based on the association relationships and association strengths recorded in the feature association matrix, the features in the initial feature set and the prior feature set are fused to generate a comprehensive feature set. Then, using the features in the comprehensive feature set as nodes, a knowledge graph is constructed based on the association relationships between the features. In the knowledge graph, nodes represent different features, and edges represent the association relationships between features.

[0098] For example, vehicle type (large truck), accident severity (involving multiple vehicles), and the relationships between them can be integrated into a single comprehensive feature. This comprehensive feature reflects the impact of large trucks on rescue efforts in multi-vehicle accident scenarios. All integrated features are then grouped into a comprehensive feature set. A knowledge graph is constructed using features from this comprehensive feature set as nodes. For instance, "impact of large trucks in multi-vehicle accidents" can be one node, and "accident severity" another node. The edges between them represent the relationships between the two (such as obstructing rescue access). In the knowledge graph, different nodes represent different comprehensive features, and the edges accurately describe the logical connections between features.

[0099] Step S134: Analyze the knowledge graph using a graph mining algorithm to mine potential knowledge patterns and relationship rules in the knowledge graph. The knowledge patterns and relationship rules are used to reflect the inherent laws and logic in the emergency response process.

[0100] For example, in the knowledge graph during the emergency response deployment phase, graph mining algorithms might uncover patterns such as: when an accident is severe and large vehicles are nearby, emergency response deployment should prioritize opening up access routes around the large vehicles to facilitate subsequent rescue vehicles. Relationship rules might suggest that if there's a specific relationship between the traffic congestion area and vehicle positions (e.g., vehicles are on the edge of the congestion area and close to the direction of rescue resource access), then vehicles should be guided in a specific direction to mitigate the impact of congestion on rescue efforts. These knowledge patterns and relationship rules are mined from the structure of the knowledge graph and the relationships between nodes, contributing to a deeper understanding of the patterns in the emergency response process.

[0101] Step S135: The knowledge patterns and relationship rules mined from the knowledge graph are screened, and knowledge patterns and relationship rules whose contribution to the generation and transmission of knowledge vector data is less than the set contribution are removed. The importance of the screened knowledge patterns and relationship rules is evaluated, and an importance weight is assigned to each knowledge pattern and relationship rule.

[0102] For example, during the rescue force deployment phase, if a discovered knowledge pattern concerns the relationship between vehicle color and rescue (assuming vehicle color has almost no impact on rescue force deployment), it will be discarded because its contribution to generating the knowledge vector data is extremely low. For the remaining knowledge patterns and relationship rules, such as "the impact of large vehicles in an accident and the relationship between rescue force deployment," their importance is assessed based on their actual role in emergency response. For instance, "the relationship between large vehicles obstructing rescue routes and the order of rescue vehicle deployment" might be assigned a higher importance weight because it directly relates to rescue efficiency.

[0103] Step S136: Based on the information recorded in the knowledge pattern and relation rule set, and according to the importance weight assigned to each knowledge pattern and relation rule, the features in the comprehensive feature set are fused and adjusted to generate an optimized comprehensive feature set. The optimized comprehensive feature set is then integrated and encoded and converted into a vector representation to generate the transfer knowledge vector data corresponding to the candidate emergency response stage.

[0104] Finally, during the rescue force deployment phase, features such as "the impact of large vehicles in multi-vehicle accidents" in the comprehensive feature set are integrated and adjusted according to the previously assigned importance weights. If the feature "obstruction of rescue channels by large vehicles" has a higher importance weight, its impact will be more prominent during the integration and adjustment process. After this processing, an optimized comprehensive feature set is obtained. Then, the features in this set are integrated and encoded according to a specific encoding method (such as based on feature type, numerical range, etc.), and transformed into vector form. This vector is the transferable knowledge vector data corresponding to the rescue force deployment phase. It contains comprehensive information about the driving events of the control unit during the rescue force deployment phase after comprehensively considering various factors, and can provide accurate data support for subsequent emergency response operations. The same process will be carried out in other emergency response phases such as traffic management and post-accident recovery, generating transferable knowledge vector data that accurately reflects the situation at each phase.

[0105] In one possible implementation, the training steps of the emergency response analysis model include:

[0106] Step S101: Obtain sample learning data, which includes the sample event logic knowledge network of sample highway emergency events and sample driving event data of sample control unit driving events.

[0107] In this embodiment, taking a sample vehicle rollover accident on a highway as an example, the sample event logic knowledge network covers multiple aspects of information. It includes the precise location of the accident site on the highway (e.g., between two specific kilometer markers and near a service area exit), the lane layout of the accident section (number of lanes in one direction, presence of emergency lanes, etc.), the distribution of rescue resources around the accident site (e.g., distance to the nearest fire station and emergency medical center, and driving routes), and the traffic flow patterns of the section (average traffic flow at different times, peak traffic flow during peak hours, etc.). The sample driving event data of the sample control unit is related to the driving events monitored by each control unit in the accident scenario. For example, if a control unit on the accident section monitors a car slowing down after the accident, its sample driving event data includes the car's speed change curve (a record of the speed gradually decreasing from normal to a stable low speed, and the time points of the speed change), the vehicle's direction of travel (whether it is traveling in the same or opposite direction, and specific lane information), and the vehicle type (small passenger car), etc. Another control unit might monitor a rescue vehicle heading towards the accident scene, and its sample driving event data includes information such as the rescue vehicle's speed, departure location, and estimated arrival time. These sample event logical knowledge networks and sample driving event data together constitute the sample learning data.

[0108] Step S102: Based on the sample event logical knowledge network and the sample driving event data, the initial logical learning module in the initialized emergency response analysis model is used to obtain the sample initial knowledge vector data of the sample control unit driving event at multiple emergency response stages, wherein the multiple emergency response stages have a logical temporal order.

[0109] Next, in the example vehicle rollover accident, the emergency response phase can be divided into the initial information collection and assessment phase, the rescue force deployment phase, the traffic control phase, and the post-accident recovery phase. In the initial information collection and assessment phase, the initial logic learning module processes information such as accident location, surrounding rescue resources, and vehicle speed and type from the example event logic knowledge network. For example, accident location information is encoded based on its coordinates on the highway, transforming it into a specific vector element; surrounding rescue resource information, such as the distance to the nearest rescue station in kilometers, is encoded into a vector according to a certain numerical mapping rule; vehicle speed is divided into different intervals based on the speed range and encoded into corresponding vector elements; and vehicle type is converted into a vector element according to a pre-set type encoding method. Through this processing, the initial logic learning module obtains the initial knowledge vector data of the example control unit's driving event in the initial information collection and assessment phase. In the rescue force deployment phase, the initial logic learning module focuses on processing information related to the deployment of rescue resources. For example, information about areas around the accident site suitable for rescue vehicles to park is extracted from the example event logical knowledge network, and information such as the departure point and speed of rescue vehicles is obtained from the example driving event data. After specific encoding and logical processing, the initial knowledge vector data for that stage is obtained. Similarly, in the traffic management stage and the recovery stage after the accident cleanup, the initial logic learning module will also process the data according to the logical relationships of each stage based on the corresponding example event logical knowledge network and example driving event data, thereby obtaining the initial knowledge vector data for each emergency response stage.

[0110] Step S103: For each emergency response stage, the initial knowledge vector data of the sample control unit driving event in the candidate emergency response stage is interactively fused with the corresponding prior sample transfer knowledge vector data through the knowledge transfer fusion module corresponding to the candidate emergency response stage in the initialized emergency response analysis model, so as to generate sample transfer knowledge vector data corresponding to the candidate emergency response stage.

[0111] Then, during the rescue force deployment phase, for a specific control unit's driving event, such as the rescue vehicles heading to the accident scene mentioned earlier, there is initial sample knowledge vector data during this phase. This data includes information such as the rescue vehicle's speed, departure point, and estimated arrival time. The corresponding prior sample transfer knowledge vector data is generated during the initial accident information collection and assessment phase and is related to the overall accident situation. This includes information such as the severity of the accident (damage to the overturned vehicle, whether there are casualties, etc.) and the traffic congestion range at the accident scene (the congestion kilometers extending forward and backward from the accident point). The knowledge transfer and fusion module first analyzes and processes these two datasets. It may generate intermediate calculation results based on factors such as the relationship between the rescue vehicle's speed and the traffic congestion range at the accident scene, and the relationship between the rescue vehicle's departure point and the distribution of rescue resources around the accident site. For example, if the rescue vehicle's speed is high and the traffic congestion range at the accident scene is small, a specific fusion coefficient may be generated. Then, based on this fusion coefficient, the initial sample knowledge vector data, and the prior sample transfer knowledge vector data, reference sample knowledge vector data is generated using a specific calculation method. Finally, based on this fusion coefficient, the initial sample knowledge vector data, and the reference sample knowledge vector data, sample transfer knowledge vector data corresponding to the rescue force deployment stage is generated. This sample transfer knowledge vector data may contain more comprehensive information about the deployment of rescue forces, such as whether it is necessary to adjust the driving routes of rescue vehicles or whether it is necessary to coordinate other rescue resources. The same process is carried out during the traffic control stage and the recovery stage after the accident cleanup, generating corresponding sample transfer knowledge vector data for each emergency response stage.

[0112] Step S104: Based on the sample knowledge vector data corresponding to each emergency response stage, determine the sample response plan corresponding to each emergency response stage.

[0113] Subsequently, in the initial information gathering and assessment phase of the accident, based on information such as the accident location, traffic congestion area, and vehicle movement status from the sample knowledge vector data, the sample response plan might involve dispatching more monitoring equipment to the vicinity of the accident scene to obtain more detailed accident information, while simultaneously monitoring vehicle speeds within a certain range around the accident scene and providing timely feedback. In the rescue force deployment phase, based on information such as rescue vehicles and the severity of the accident from the sample knowledge vector data, the sample response plan might involve adjusting the routes of some rescue vehicles, prioritizing the dispatch of rescue vehicles that are closer to the accident scene and are fully equipped, while coordinating surrounding traffic control to ensure priority passage for rescue vehicles. In the traffic management phase, based on information such as traffic congestion and vehicle flow from the sample knowledge vector data, the sample response plan might involve setting up reasonable diversion points to guide vehicles to detour in an orderly manner, and developing different management strategies based on the characteristics of different types of vehicles (such as buses and trucks). During the recovery phase after the accident cleanup, based on information such as the damage to road facilities and the driving status of remaining vehicles from the sample knowledge vector data, the sample response plan may be to quickly repair and inspect the damaged road facilities, gradually restore normal traffic control measures, and guide vehicles to resume normal driving speeds.

[0114] Step S105: Determine the training error parameters based on the sample response schemes and labeled response schemes corresponding to each emergency response stage.

[0115] The labeled response plan data consists of pre-defined ideal response plans for each stage of an emergency response. In the initial information gathering and assessment phase, the labeled response plan may specify the acquisition of certain key information (such as precise casualty figures and detailed vehicle damage) within a specific timeframe, and the completion of traffic flow monitoring in a specific area. Comparing the sample response plan with the labeled response plan data will introduce errors if the sample response plan takes longer to acquire key information than the labeled response plan specifies, or if traffic flow monitoring in a specific area is not completed. In the rescue force deployment phase, the labeled response plan may define the optimal arrival time of rescue vehicles at the accident scene and the best allocation method for rescue resources. Errors will also occur if the arrival time of rescue vehicles in the sample response plan is later than the optimal time in the labeled response plan, or if the allocation method for rescue resources is not optimal. By performing such comparative analysis of the sample response plan and labeled response plan data for each emergency response stage, the error value for each stage is calculated, and then the overall training error parameters are determined based on these error values.

[0116] Step S106: Train the initialized emergency response analysis model according to the training error parameters until the initialized emergency response analysis model meets the model convergence condition, and generate the completed emergency response analysis model.

[0117] Finally, during training, if the training error parameter is large, it indicates a significant discrepancy between the emergency response analysis model's predictions and the ideal labeled response plan data. In this case, the model will adjust its internal parameters. For example, in the initial logic learning module, it will adjust the encoding methods and weight allocations for different information; in the knowledge transfer and fusion module, it will adjust the calculation methods for data fusion and the coefficient generation rules. With continuous training, the training error parameter will gradually decrease. When the training error parameter reaches a pre-set minimum value or meets other model convergence conditions (such as the change in error parameter being less than a certain threshold after multiple consecutive training iterations), the initialized emergency response analysis model is considered to have been successfully trained, thus generating an emergency response analysis model capable of accurately performing emergency response analysis. This successfully trained emergency response analysis model can be applied to actual highway emergency event handling, accurately generating response plans for each emergency response stage based on the input event logic knowledge network and control unit driving event data.

[0118] In one possible implementation, prior to step S110, the method further includes:

[0119] Based on the activation command of the emergency response system, an emergency event tag is extracted from the activation command and the emergency event tag is used as a highway emergency event.

[0120] After step S150, the method further includes: based on the activation operation of the emergency response interface, presenting the response events corresponding to the highway emergency events in order of priority.

[0121] In this embodiment, when the emergency response system receives a activation command, this command contains key information about the emergency event. For example, the activation command may be issued by staff at the monitoring center after discovering an anomaly on the highway, or it may be triggered by automatic monitoring equipment (such as smart cameras, vehicle sensor networks, etc.) when it detects a specific situation.

[0122] Suppose the emergency response system receives an activation command from monitoring center staff. This command is issued after staff observe a large number of vehicles suddenly slowing down and smoke being produced on a section of highway via surveillance footage. The activation command contains information related to the event in a specific data format, such as the approximate area of ​​the event (identified as the section between two highway exits based on location information from the surveillance footage), the initial characteristics of the event (vehicle slowdown and smoke), and the time of the event.

[0123] The emergency response system begins extracting emergency event tags from the activation command. An emergency event tag is a summary identifier for the entire emergency event, extracted from the numerous pieces of information in the activation command. In the example above, the emergency event tag might be defined as "an event of abnormal vehicle deceleration and smoke on a section of road between exits." This tag accurately describes the key characteristics and location of the event, facilitating subsequent emergency handling and management. This emergency event tag is then used as the core identifier for the entire emergency response process on the highway. All subsequent operations, such as acquiring the event logic knowledge network, managing traffic event data from control units, and determining response plans, revolve around this highway emergency event.

[0124] In the process of handling highway emergencies, the emergency response interface is a crucial interface for staff to interact with and manage the emergency. Staff will activate the emergency response interface when they need to handle a highway emergency.

[0125] Suppose that during the emergency response to the previously mentioned incident of "vehicles decelerating abnormally and smoke appearing on a section of road between exits," the emergency response interface is activated. This interface contains multiple functional modules and information display areas, specifically designed for handling highway emergency events.

[0126] Based on the previously determined priority information for control unit driving events, response events will be presented in priority order. The response event corresponding to the highest priority control unit driving event will be displayed first. For example, if previous analysis determined that the driving event of a rescue vehicle carrying firefighting equipment and heading towards a smoke-generating area is of high priority, then the response event corresponding to this driving event will be presented first. This response event may include the optimal driving route planned for the rescue vehicle (displayed on the map module of the emergency response interface, with highlighted lines indicating the lane and direction the rescue vehicle should travel), the estimated arrival time of the rescue vehicle (calculated by the rescue vehicle's current speed, distance traveled, and road conditions, and displayed numerically on the interface), and precautions that the rescue vehicle needs to take during its journey (such as temporary traffic control measures that may exist on certain road sections, which will be displayed in text form on the interface).

[0127] Next, the corresponding handling events for the next-priority control unit driving events will be displayed sequentially. For example, if a control unit monitors a vehicle illegally changing lanes and attempting to cut in front of an accident section, and this driving event is determined to be of next-priority, then the emergency response interface will display the handling events for this illegally changing vehicle. These events include dispatching traffic enforcement officers to stop the violation (displaying the dispatch information of the officers on the interface, such as which enforcement station they will depart from), penalty suggestions for the driver of the violating vehicle (listing possible penalty clauses on the interface according to relevant traffic regulations), and how to guide the violating vehicle to a safe area to await further processing (displaying the guidance route on the interface).

[0128] As the priority decreases, the corresponding response events for driving events in other control units will be presented on the emergency response interface in sequence. For example, for vehicles driving normally around the accident section, their driving events have a lower priority. The corresponding response event might be to send a notification to these vehicles through the traffic broadcast system, informing them of the event ahead and precautions to take (such as slowing down and maintaining a safe distance). The emergency response interface will display information such as the broadcast status (sent or not sent), the broadcast content, and the expected range of vehicles covered.

[0129] By presenting the corresponding emergency events on the emergency response interface in this priority order, staff can clearly and systematically understand the handling status of driving events in each control unit, thereby efficiently handling and managing emergency events and ensuring the safety and smooth flow of traffic on the highway.

[0130] Figure 2 The diagram illustrates the hardware structure of a highway emergency response system 100 based on a control unit, provided by an embodiment of the present invention, for implementing the aforementioned highway emergency response method based on a control unit. Figure 2 As shown, the highway emergency response system 100 based on the control unit may include a processor 110, a machine-readable storage medium 120, a bus 130, and a communication unit 140.

[0131] Machine-readable storage medium 120 may store data and / or instructions. In some embodiments, machine-readable storage medium 120 may store data acquired from an external terminal. In some embodiments, machine-readable storage medium 120 may store data and / or instructions used by the highway emergency response system 100 based on the control unit to perform or use in order to accomplish the exemplary methods described in this invention.

[0132] In the specific implementation process, one or more processors 110 execute computer-executable instructions stored in machine-readable storage medium 120, so that processor 110 can execute the highway emergency response method based on control unit as described in the above method embodiment. Processor 110, machine-readable storage medium 120 and communication unit 140 are connected through bus 130. Processor 110 can be used to control the sending and receiving actions of communication unit 140.

[0133] The specific implementation process of processor 110 can be found in the various method embodiments executed by the highway emergency response system 100 based on the control unit described above. The implementation principle and technical effect are similar, and will not be repeated here.

[0134] Furthermore, this embodiment of the invention also provides a readable storage medium, wherein computer-executable instructions are preset in the readable storage medium, and when the processor executes the computer-executable instructions, the above-mentioned highway emergency response method based on the control unit is implemented.

[0135] It should be noted that, in order to simplify the description of the present invention and thus help to understand one or more embodiments of the invention, multiple features may sometimes be grouped into one embodiment, drawing or description thereof in the foregoing description of the embodiments of the present invention.

Claims

1. A method for emergency response on a highway based on a control unit, characterized in that, The method comprises: acquiring an event logical knowledge network of a highway emergency event and driving event data corresponding to each driving event of a control unit in a control unit monitoring stream of the highway emergency event; acquiring initial knowledge vector data of each driving event of the control unit in multiple emergency response stages according to the event logical knowledge network and the driving event data corresponding to each driving event of the control unit, wherein the multiple emergency response stages have a logical time sequence order; for each driving event of the control unit, interactively fusing the initial knowledge vector data of the candidate driving event of the control unit in each emergency response stage and the corresponding prior transfer knowledge vector data to generate transfer knowledge vector data corresponding to each emergency response stage, wherein the prior transfer knowledge vector data is the transfer knowledge vector data corresponding to the emergency response stages having a logical time sequence connection with the corresponding emergency response stage; determining response schemes corresponding to each emergency response stage according to the transfer knowledge vector data corresponding to each emergency response stage, and determining priority information of the candidate driving event of the control unit according to the response schemes corresponding to each emergency response stage; determining a disposal event corresponding to the highway emergency event according to the priority information corresponding to each driving event of the control unit in the control unit monitoring stream; the step of acquiring the initial knowledge vector data of each driving event of the control unit in the multiple emergency response stages according to the event logical knowledge network and the driving event data corresponding to each driving event of the control unit comprises: for each driving event of the control unit, acquiring the initial knowledge vector data of the candidate driving event of the control unit in the multiple emergency response stages by an initial logical learning module in an emergency response analysis model according to the event logical knowledge network and the driving event data corresponding to the candidate driving event of the control unit; the step of interactively fusing, for each driving event of the control unit, the initial knowledge vector data of the candidate driving event of the control unit in each emergency response stage and the corresponding prior transfer knowledge vector data to generate transfer knowledge vector data corresponding to each emergency response stage comprises: for each emergency response stage, interactively fusing, by a knowledge transfer fusion module corresponding to the candidate emergency response stage in the emergency response analysis model, the initial knowledge vector data of the candidate driving event of the control unit in the candidate emergency response stage and the corresponding prior transfer knowledge vector data to generate transfer knowledge vector data corresponding to the candidate emergency response stage; the step of interactively fusing, by the knowledge transfer fusion module corresponding to the candidate emergency response stage in the emergency response analysis model, the initial knowledge vector data of the candidate driving event of the control unit in the candidate emergency response stage and the corresponding prior transfer knowledge vector data to generate transfer knowledge vector data corresponding to the candidate emergency response stage comprises: normalizing the initial knowledge vector data and the prior transmission knowledge vector data, extracting first feature information related to emergency response from the normalized initial knowledge vector data to generate an initial feature set, the first feature information including event type features, event severity features, event occurrence location features, and extracting second feature information associated with emergency response history from the normalized prior transmission knowledge vector data to generate a prior feature set, the second feature information including processing mode features of historical similar events, influence features of historical events on the current event; performing matching analysis on the features in the initial feature set and the prior feature set to determine feature pairs that are associated with each other in the initial feature set and the prior feature set, analyzing the internal logical relationship of a target feature pair with an association degree greater than a set threshold, and constructing a feature association matrix according to the internal logical relationship analysis result, the feature association matrix being used to record the association relationship and association strength between features; fusing the features in the initial feature set and the prior feature set according to the association relationship and association strength recorded in the feature association matrix to generate a comprehensive feature set, and constructing a knowledge graph according to the association relationship between features with the features in the comprehensive feature set as nodes, in which nodes represent different features and edges represent the association relationship between features; analyzing the knowledge graph using a graph mining algorithm to mine potential knowledge patterns and relationship rules in the knowledge graph, the knowledge patterns and relationship rules being used to reflect the internal laws and logic in the emergency response process; screening the knowledge patterns and relationship rules mined in the knowledge graph, eliminating knowledge patterns and relationship rules with a contribution degree to generating transmission knowledge vector data less than a set contribution degree, and performing importance evaluation on the screened knowledge patterns and relationship rules to assign an importance weight to each knowledge pattern and relationship rule; fusing and adjusting the features in the comprehensive feature set according to the information recorded in the knowledge pattern and relationship rule set and the importance weight assigned to each knowledge pattern and relationship rule to generate an optimized comprehensive feature set, and integrating and encoding the optimized comprehensive feature set to convert it into a vector form to generate transmission knowledge vector data corresponding to a candidate emergency response stage.

2. The tube and lens cell based highway emergency response method according to claim 1, wherein, The initial knowledge vector data of each control unit driving event in multiple emergency response stages is obtained according to the event logic knowledge network and the driving event data corresponding to each control unit driving event, including: For each control unit driving event, knowledge encoding representation is performed on the event logic knowledge network, the driving event data corresponding to the candidate control unit driving event, and the emergency response condition information to generate each knowledge encoding vector; The knowledge encoding vectors are integrated to generate integrated knowledge encoding data; The integrated knowledge encoding data is logically reconstructed by each logical learning module to generate logical reconstruction data corresponding to each logical learning module. Obtaining the weighting coefficients corresponding to each logical learning module respectively; Based on the target logical reconstruction data and the target weighting coefficients corresponding to each target logical learning module corresponding to each emergency response stage, determine the fusion knowledge vector data corresponding to each emergency response stage respectively; Self-attention processing is performed on the fusion knowledge vector data corresponding to each emergency response stage respectively to generate initial knowledge vector data of the candidate management and control unit driving event in each emergency response stage.

3. The tube and lens cell based highway emergency response method of claim 2, wherein, The acquisition of the weighting coefficients corresponding to each logical learning module respectively includes: According to the importance space array, the target conversion data matching the number of logical learning modules is generated by space conversion; The target conversion data is stimulated to generate the weighting coefficients corresponding to each logical learning module respectively.

4. The tube and lens cell based highway emergency response method of claim 2, wherein, The determination of the fusion knowledge vector data corresponding to each emergency response stage based on the target logical reconstruction data and the target weighting coefficients corresponding to each target logical learning module corresponding to each emergency response stage includes: For each emergency response stage, the target logical reconstruction data and the corresponding target weighting coefficients of each target logical learning module corresponding to the candidate emergency response stage are weighted to generate the weighted calculation result of each target logical learning module corresponding to the candidate emergency response stage. The weighted calculation results of each target logical learning module corresponding to the candidate emergency response stage are fused to generate the fusion knowledge vector data corresponding to the candidate emergency response stage.

5. The tube and lens cell based highway emergency response method of claim 1, wherein, The interactive fusion of the initial knowledge vector data of the candidate management and control unit driving event in each emergency response stage and the corresponding prior transfer knowledge vector data to generate the transfer knowledge vector data corresponding to each emergency response stage includes: For each emergency response stage, a knowledge fusion coefficient is generated according to the initial knowledge vector data of the candidate management and control unit driving event in the candidate emergency response stage and the corresponding prior transfer knowledge vector data; According to the initial knowledge vector data of the candidate management and control unit driving event in the candidate emergency response stage, the corresponding prior transfer knowledge vector data, and the knowledge fusion coefficient, a reference knowledge vector data is generated; According to the knowledge fusion coefficient, the initial knowledge vector data of the candidate management and control unit driving event in the candidate emergency response stage, and the reference knowledge vector data, the transfer knowledge vector data corresponding to the candidate emergency response stage is generated.

6. The tube and lens cell based highway emergency response method of claim 1, wherein, The determination of the priority information of the candidate management and control unit driving event according to the response scheme corresponding to each emergency response stage includes: Global analysis is performed on the response scheme corresponding to each emergency response stage to generate the priority information of the candidate management and control unit driving event.

7. The tube and lens cell based highway emergency response method of claim 1, wherein, The training steps of the emergency response analysis model include: Obtaining sample learning data, the sample learning data including sample event logical knowledge network of sample expressway emergency event and sample driving event data of sample management and control unit driving event; According to the sample event logic knowledge network and the sample driving event data, through the initial logic learning module in the initialized emergency response analysis model, sample initial knowledge vector data of the sample management and control unit driving event in multiple emergency response stages is obtained, and the multiple emergency response stages have a logical time sequence order; For each emergency response stage, through the knowledge transfer fusion module corresponding to the candidate emergency response stage in the initialized emergency response analysis model, the sample initial knowledge vector data of the sample management and control unit driving event in the candidate emergency response stage and the corresponding prior sample transfer knowledge vector data are interactively fused to generate sample transfer knowledge vector data corresponding to the candidate emergency response stage; According to the sample transfer knowledge vector data corresponding to each emergency response stage, sample response schemes corresponding to each emergency response stage are determined; According to the sample response schemes corresponding to each emergency response stage and the labeled response scheme data corresponding to each emergency response stage, a training error parameter is determined; According to the training error parameter, the initialized emergency response analysis model is trained until the initialized emergency response analysis model meets the model convergence condition, and a trained emergency response analysis model is generated.

8. The tube and lens cell based highway emergency response method of claim 1, wherein, Before the step of obtaining the event logic knowledge network of the highway emergency event and the driving event data corresponding to each management and control unit driving event in the management and control unit monitoring flow, the method further comprises: According to the start instruction of the emergency response system, an emergency event label is extracted from the start instruction, and the emergency event label is used as a highway emergency event; After the step of determining the disposal event corresponding to the highway emergency event according to the priority information corresponding to each management and control unit driving event in the management and control unit monitoring flow, the method further comprises: According to the activation operation of the emergency disposal interface, the disposal events corresponding to the highway emergency event are respectively presented in priority order.

9. A tube and management unit based highway emergency response system, characterized in that, The highway emergency response system based on the management and control unit comprises a processor and a memory, the memory and the processor are connected, the memory is used to store programs, instructions or codes, and the processor is used to execute the programs, instructions or codes in the memory to realize the highway emergency response method based on the management and control unit in any one of claims 1-8.

Images