Automatic scheduling method for road obstacle clearance rescue enterprise

Abstract

The invention relates to the technical field of traffic management, and discloses a road obstacle clearance rescue enterprise automatic scheduling method comprising the following steps: S1, through multi-source data fusion and a dynamic convolutional neural network, generating a three-dimensional road model, and predicting accidents and congestion; s2, dynamically adjusting the priority of the rescue task by using a multi-target genetic algorithm in combination with real-time data; s3, path planning and cooperative scheduling are optimized through deep reinforcement learning and game theory algorithms; s4, dynamically correcting the path by using a vehicle-mounted sensor, and optimizing the flow and the energy consumption; s5, in an emergency, an emergency module based on reinforcement learning starts a plan and adjusts resources; and S6, after the task, analyzing an optimization strategy through collaborative learning, and carrying out self-improvement. According to the method, multi-source dynamic data fusion is carried out, a dynamic convolutional neural network is used for processing data in real time, a high-precision three-dimensional road environment model is generated, accidents and congestion are predicted, and it is ensured that a rapid and accurate scheduling decision can be made in a complex and changeable road environment.

Description

Technical Field

[0001] The present invention relates to the technical field of traffic management, and in particular to an automatic dispatching method for a road obstacle removal and rescue enterprise. Background Art

[0002] Road clearance and rescue missions refer to dispatching professional rescue vehicles and personnel to the scene of traffic accidents, vehicle breakdowns or other situations that lead to road blockages to conduct rapid and effective cleanup and rescue in order to restore road traffic capacity. Such missions usually include clearing the accident site, towing of faulty vehicles, traffic diversion, and rescue of trapped vehicles or personnel. Road clearance and rescue missions play an important role in maintaining traffic order, reducing traffic congestion, and ensuring road safety. Especially in urban traffic, timely and effective clearance and rescue can significantly reduce the secondary risks and economic losses caused by accidents.

[0003] With the acceleration of urbanization and the continuous increase in traffic volume, the importance of road clearance and rescue tasks in daily traffic management has become increasingly prominent; however, traditional road clearance and rescue dispatch systems usually rely on manual judgment or static rule configuration, lacking the ability to perceive and dynamically respond to real-time complex road environments. This dispatch method is often difficult to respond quickly and accurately when faced with sudden accidents, multi-vehicle chain rear-end collisions, or traffic jams caused by extreme weather, resulting in long response times for rescue tasks and unreasonable resource allocation, which in turn affects the operating efficiency of the entire transportation system;

[0004] In the existing technology, although some dispatching systems have begun to introduce data fusion and path planning technologies, most systems still have problems such as single data processing, insufficient intelligence in path planning, and delayed emergency response. First, traditional systems usually rely on a single or small number of data sources and cannot fully integrate information from multiple data sources such as traffic management systems, meteorological information, and Internet of Vehicles data, resulting in insufficient perception of complex road environments. In addition, existing systems often use static rules or simple optimization algorithms in task priority evaluation and path planning, lacking the ability to dynamically adjust real-time data, and are unable to cope with the rapid changes in road conditions and the complex needs of multi-task scheduling.

[0005] Especially when extreme weather or large-scale traffic accidents occur, the existing system has limited emergency response capabilities and is unable to quickly initiate emergency plans and reasonably allocate resources. This leads to inefficient execution of rescue missions and may even delay rescue time due to improper resource scheduling. At the same time, the existing system lacks an effective feedback and learning mechanism after the mission is completed, and is unable to accumulate experience and perform self-optimization from each rescue mission. Therefore, it is difficult to improve the accuracy and flexibility of scheduling strategies in subsequent tasks. Summary of the invention

[0006] In view of the deficiencies in the prior art, the present invention provides an automatic dispatching method for road obstacle removal and rescue enterprises, which solves the problem that dispatching decisions are not fast enough.

[0007] To achieve the above objectives, the present invention is implemented through the following technical solutions: a method for automatically dispatching a road obstacle removal and rescue enterprise, comprising the following steps:

[0008] S1. Multi-source dynamic data fusion and deep environmental perception: Through the multi-source dynamic data fusion module, multi-source data such as traffic management system, meteorological information, Internet of Vehicles data, drone monitoring, smart cameras and mobile user reports are collected in real time, and dynamic convolutional neural network is used for deep environmental perception to generate a high-precision three-dimensional road environment model for predicting accidents and congestion.

[0009] S2. Task priority evaluation based on multi-dimensional optimization: Combined with road conditions, weather conditions, accident severity and rescue vehicle status, a multi-dimensional optimization evaluation method based on a multi-objective genetic algorithm is used to dynamically adjust the priority of rescue tasks, and a weighted adaptive learning mechanism is used to optimize the task priority sorting strategy in real time;

[0010] S3. Adaptive path planning and dynamic multi-agent scheduling: Through the multi-agent deep reinforcement learning algorithm, the system learns and optimizes path planning in real time, taking into account the real-time location of the vehicle, traffic flow, road condition changes and the urgency of the rescue mission, to achieve adaptive path planning. At the same time, the dynamic game theory algorithm is used to coordinate the coordinated scheduling of multiple rescue vehicles to ensure the overall optimality of resource allocation and path planning.

[0011] S4, Real-time road environment feedback and multi-dimensional path optimization: During the execution of the system, the system obtains feedback information about the road environment in real time through on-board sensors and edge computing nodes, and uses a multi-dimensional path optimization algorithm to dynamically correct the path. The algorithm combines traffic flow prediction, vehicle energy consumption optimization, and emergency lane usage strategy to achieve optimal path planning for rescue vehicles;

[0012] S5. Intelligent emergency response and dynamic re-dispatch: In the event of an emergency or extreme weather, the system automatically initiates the emergency plan through the emergency response module based on reinforcement learning, dynamically adjusts rescue resources and dispatch strategies, and ensures the priority execution of rescue tasks. The module can reallocate resources and reallocate rescue tasks in real time according to changes in road conditions to cope with multiple rescue needs.

[0013] S6. Dynamic collaborative learning and post-mission evaluation: After the rescue mission is completed, the system uses a dynamic collaborative learning mechanism to analyze the path selection, task allocation and resource usage during the rescue process, generate a detailed mission report and automatically update the path planning model and scheduling strategy to achieve self-optimization and continuous improvement of the system.

[0014] Preferably, in step S1, the algorithm used by the dynamic convolutional neural network is:

[0015] y=f(W*X+b)

[0016] Among them, y is the output feature map, W is the convolution kernel, X is the input data, b is the bias term, * is the convolution operation, and f is the activation function.

[0017] Preferably, in step S1, the three-dimensional road environment model establishment parameters are:

[0018] Input dimension: feature dimension of multi-source data (such as image data, sensor data);

[0019] Convolution kernel size: usually set to 3x3 or 5x5, depending on the data characteristics;

[0020] Number of layers: The number of DCNN layers is determined by model complexity and task requirements, usually 3-10 layers;

[0021] The steps of establishing the three-dimensional road environment model are:

[0022] a. Data preprocessing: standardize and normalize multi-source data;

[0023] b. Convolution operation: extract features from data through multi-layer convolution, and obtain feature maps at different levels layer by layer;

[0024] c. 3D reconstruction: Use convolution features to generate a high-precision 3D road environment model;

[0025] The multi-source data are all acquired using the latest information.

[0026] Preferably, in step S2, the multi-objective genetic algorithm formula is:

[0027]

[0028] F i is the fitness value of the i-th task, w j is the weight of the j-th dimension target, dynamically adjusted, f j (x i ) is the j-th dimension objective function, x i is the feature vector of the i-th task.

[0029] Preferably, in step S3, multi-agent deep reinforcement learning is combined with a dynamic game theory algorithm, and the corresponding algorithm formula is:

[0030]

[0031] Among them, Q(s,a) is the value of taking action a in state s, α is the learning rate, r is the immediate reward, γ is the discount factor, weighing future rewards, s′ is the next state, is the maximum expected return in the next state;

[0032] When the intelligent agent dispatching system detects a failure of a rescue vehicle, it can automatically reallocate tasks and adjust the paths of other vehicles.

[0033] Preferably, in step S4, the multi-dimensional path optimization algorithm formula is:

[0034]

[0035] Among them, L total is the total path cost, w k is the weight of the k-th dimension path factor, L k is the k-th dimension path cost;

[0036] The dynamic correction is performed sequentially according to the urgency and reliability of the feedback.

[0037] Preferably, in step S5, the dynamic adjustment is determined according to a policy gradient algorithm, and the policy gradient algorithm formula is:

[0038]

[0039] Among them, J(θ) is the expected return of the strategy, π θ (a|s) is the probability of taking action a in state s, is the value function;

[0040] The dispatching resources will reserve a certain proportion of rescue resources to deal with emergencies.

[0041] Preferably, in step S6, the formula used by the dynamic collaborative learning mechanism is:

[0042]

[0043] Among them, Δθ is the update value of the model parameters, η is the learning rate, (y i ,f(x i ; θ)) is the task loss function, x i and i is the data and results of the i-th task.

[0044] The present invention provides an automatic dispatching method for road obstacle removal and rescue enterprises. It has the following beneficial effects:

[0045] 1. The present invention uses multi-source dynamic data fusion and deep environmental perception technology, and utilizes dynamic convolutional neural networks to process multiple data sources such as traffic management systems, meteorological information, and vehicle networking data in real time, and generates a high-precision three-dimensional road environment model. This model can accurately predict the occurrence of accidents and congestion, ensure that the dispatching system can make fast and accurate decisions in complex and changeable road environments, and effectively improve the response speed and execution efficiency of rescue missions.

[0046] 2. The scheduling optimization mechanism of the present invention is based on a multi-objective genetic algorithm and a multi-agent deep reinforcement learning algorithm. In the process of task priority evaluation and path planning, dynamic adaptive resource scheduling is realized. By dynamically adjusting task weights and optimizing the paths of rescue vehicles, the system can automatically reallocate tasks in emergencies to ensure maximum resource utilization and smooth completion of rescue missions, greatly improving the flexibility and reliability of the scheduling system.

[0047] 3. The present invention introduces intelligent emergency response and dynamic collaborative learning mechanisms. Through the emergency response module based on the policy gradient algorithm and the dynamic re-scheduling strategy, the system can automatically initiate emergency plans under extreme weather or complex traffic conditions and reserve rescue resources to deal with emergencies. In addition, the dynamic collaborative learning mechanism after the task is completed enables the system to continuously optimize the scheduling strategy and path planning model, ensuring that the scheduling system can improve itself after each task, thereby improving the execution effect of future rescue tasks. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 The figure is a flow chart of the method steps of the present invention. DETAILED DESCRIPTION

[0049] The following will be combined with the drawings in the specification of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0050] Embodiment 1:

[0051] Please refer to the attached Figure 1 The embodiment of the present invention provides a method for automatically dispatching a road obstacle removal and rescue enterprise, comprising the following steps:

[0052] S1. Multi-source dynamic data fusion and deep environmental perception: Through the multi-source dynamic data fusion module, multi-source data such as traffic management system, meteorological information, Internet of Vehicles data, drone monitoring, smart cameras and mobile user reports are collected in real time, and dynamic convolutional neural network is used for deep environmental perception to generate a high-precision three-dimensional road environment model for predicting accidents and congestion.

[0053] S2. Task priority evaluation based on multi-dimensional optimization: Combined with road conditions, weather conditions, accident severity and rescue vehicle status, a multi-dimensional optimization evaluation method based on a multi-objective genetic algorithm is used to dynamically adjust the priority of rescue tasks, and a weighted adaptive learning mechanism is used to optimize the task priority sorting strategy in real time;

[0054] S3. Adaptive path planning and dynamic multi-agent scheduling: Through the multi-agent deep reinforcement learning algorithm, the system learns and optimizes path planning in real time, taking into account the real-time location of the vehicle, traffic flow, road condition changes and the urgency of the rescue mission, to achieve adaptive path planning. At the same time, the dynamic game theory algorithm is used to coordinate the coordinated scheduling of multiple rescue vehicles to ensure the overall optimality of resource allocation and path planning.

[0055] S4, Real-time road environment feedback and multi-dimensional path optimization: During the execution of the system, the system obtains feedback information about the road environment in real time through on-board sensors and edge computing nodes, and uses a multi-dimensional path optimization algorithm to dynamically correct the path. The algorithm combines traffic flow prediction, vehicle energy consumption optimization, and emergency lane usage strategy to achieve optimal path planning for rescue vehicles;

[0056] S5. Intelligent emergency response and dynamic re-dispatch: In the event of an emergency or extreme weather, the system automatically initiates the emergency plan through the emergency response module based on reinforcement learning, dynamically adjusts rescue resources and dispatch strategies, and ensures the priority execution of rescue tasks. The module can reallocate resources and reallocate rescue tasks in real time according to changes in road conditions to cope with multiple rescue needs.

[0057] S6. Dynamic collaborative learning and post-task evaluation: After the rescue mission is completed, the system uses a dynamic collaborative learning mechanism to analyze the path selection, task allocation and resource usage during the rescue process, generate a detailed task report and automatically update the path planning model and scheduling strategy to achieve self-optimization and continuous improvement of the system;

[0058] In step S1, the algorithm used by the dynamic convolutional neural network is:

[0059] y=f(W*X+b)

[0060] Among them, y is the output feature map, W is the convolution kernel, X is the input data, b is the bias term, * is the convolution operation, and f is the activation function;

[0061] In step S1, the parameters for establishing the 3D road environment model are:

[0062] Input dimension: feature dimension of multi-source data (such as image data, sensor data);

[0063] Convolution kernel size: usually set to 3x3 or 5x5, depending on the data characteristics;

[0064] Number of layers: The number of DCNN layers is determined by model complexity and task requirements, usually 3-10 layers;

[0065] The steps to establish the 3D road environment model are as follows:

[0066] a. Data preprocessing: standardize and normalize multi-source data;

[0067] b. Convolution operation: extract features from data through multi-layer convolution, and obtain feature maps at different levels layer by layer;

[0068] c. 3D reconstruction: Use convolution features to generate a high-precision 3D road environment model;

[0069] Multi-source data are obtained using the latest information;

[0070] In step S2, the multi-objective genetic algorithm formula is:

[0071]

[0072] F i is the fitness value of the i-th task, w j is the weight of the j-th dimension target, dynamically adjusted, f j (x i ) is the j-th dimension objective function, x i is the feature vector of the i-th task;

[0073] In step S3, multi-agent deep reinforcement learning is combined with the dynamic game theory algorithm. The corresponding algorithm formula is:

[0074]

[0075] Among them, Q(s,a) is the value of taking action a in state s, α is the learning rate, r is the immediate reward, γ is the discount factor, weighing future rewards, s′ is the next state, is the maximum expected return in the next state;

[0076] When the agent dispatch system detects a failure in a rescue vehicle, it can automatically reallocate tasks and adjust the paths of other vehicles.

[0077] In step S4, the multidimensional path optimization algorithm formula is:

[0078]

[0079] Among them, L total is the total path cost, w k is the weight of the k-th dimension path factor, L k is the k-th dimension path cost;

[0080] Dynamic corrections are performed sequentially based on the urgency and reliability of the feedback;

[0081] In step S5, dynamic adjustment is determined according to the policy gradient algorithm. The formula of the policy gradient algorithm is:

[0082]

[0083] Among them, J(θ) is the expected return of the strategy, π θ (a|s) is the probability of taking action a in state s, is the value function;

[0084] The dispatching resources will reserve a certain percentage of rescue resources to deal with emergencies;

[0085] In step S6, the formula used by the dynamic collaborative learning mechanism is:

[0086]

[0087] Among them, Δθ is the update value of the model parameters, η is the learning rate, (y i ,f(x i ; θ)) is the task loss function, x i and i is the data and results of the i-th task.

[0088] Specifically, in step S1, by integrating data from multiple sources, such as traffic management systems, meteorological information, Internet of Vehicles data, drone monitoring, smart cameras, and mobile user reports, the system can perceive changes in the road environment in real time. The application of dynamic convolutional neural networks (DCNN) enables the system to extract high-dimensional features from these multi-source data and generate an accurate three-dimensional road environment model. This model can not only predict the occurrence of accidents and traffic congestion, but also provide a reliable environmental data basis for subsequent task priority evaluation and path planning. Due to the use of the latest multi-source data, the model always reflects the current road conditions, ensuring the timeliness of decision-making. and accuracy; in step S2, the task priority evaluation method based on multi-dimensional optimization can dynamically adjust the priority of the rescue task based on road conditions, weather conditions, accident severity and rescue vehicle status. The multi-objective genetic algorithm (MOGA) plays a core role in this step. By dynamically adjusting the weights of each dimension, the system can flexibly respond to various emergencies in a real-time environment to ensure that the most urgent and important rescue tasks are given priority. This evaluation mechanism greatly improves the efficiency and accuracy of task scheduling; in step S3, through the combination of multi-agent deep reinforcement learning and dynamic game theory algorithm, the system can In the process of real-time optimization of path planning and scheduling strategies, adaptive path planning enables rescue vehicles to dynamically adjust the driving path according to real-time traffic flow, road condition changes and mission urgency to ensure that they reach the destination as quickly as possible. When a rescue vehicle breaks down, the system can automatically reallocate tasks and adjust the paths of other vehicles to ensure that the overall rescue mission is not affected. The application of the multi-agent scheduling system achieves the overall optimality of resource allocation and improves rescue efficiency. In step S4, feedback information on the road environment is obtained in real time through on-board sensors and edge computing nodes, and the path is dynamically corrected using a multi-dimensional path optimization algorithm. According to the urgency and reliability of the feedback, the system can perform path correction in sequence to ensure that the rescue vehicle always travels on the optimal path. This mechanism significantly improves the response speed of the rescue mission and reduces delays caused by changes in traffic conditions. In step S5, in the event of an emergency or extreme weather, the system automatically starts the emergency plan and dynamically adjusts the rescue resources and scheduling strategy through the emergency response module based on reinforcement learning. Through the policy gradient algorithm, the system can quickly judge and adjust the emergency strategy to ensure that the rescue mission is executed first. At the same time, the system reserves a certain proportion of rescue resources for handling emergencies, which increases the flexibility and reliability of emergency scheduling.In step S6, after the rescue mission is completed, the system analyzes the path selection, task allocation and resource usage during the rescue process through a dynamic collaborative learning mechanism, and generates a detailed mission report. Through these analyses, the system can automatically update the path planning model and scheduling strategy to achieve self-optimization and continuous improvement. This mechanism ensures that the system can learn and improve after each mission, gradually improving the efficiency and accuracy of scheduling. ;

[0089] Example 2: Multi-vehicle rescue dispatch system based on intelligent emergency response

[0090] background:

[0091] On a highway in a large city, due to sudden extreme weather (such as waterlogging and reduced visibility caused by heavy rain), a multi-vehicle chain rear-end collision occurred. The accident was large in scope, resulting in multiple lanes being blocked, and emergency rescue became a top priority. Traditional rescue dispatch methods are unable to handle complex traffic conditions in real time, often resulting in delays in rescue vehicles on the way, affecting rescue efficiency.

[0092] Implementation process:

[0093] S1. Multi-source dynamic data fusion and deep environmental perception: When an accident occurs, the system starts immediately to collect real-time multi-source information, including real-time road conditions released by the traffic management system, weather warning information from the meteorological department, drone surveillance videos near the accident site, and Internet of Vehicles data from surrounding vehicles. The system processes this data through a dynamic convolutional neural network (DCNN) to quickly generate a three-dimensional road environment model of the accident site, accurately locate the accident point, identify closed lanes and passable areas, and predict the further development of road congestion.

[0094] S2. Task priority assessment based on multi-dimensional optimization: The system uses a multi-objective genetic algorithm (MOGA) to assess the priority of each rescue task based on the road closure, the speed of weather deterioration, the severity of the accident (such as casualties), and the real-time location and status of the rescue vehicle. During the assessment process, the system dynamically adjusts the weights to respond to possible weather deterioration and changes in traffic conditions to ensure that the most important rescue tasks are executed first.

[0095] S3. Adaptive path planning and dynamic multi-agent scheduling: The system activates multiple rescue vehicles based on the priority evaluation results and plans the optimal path for each vehicle through a multi-agent deep reinforcement learning algorithm. Due to the diversity of accident points and complex road conditions, the system also uses a dynamic game theory algorithm to coordinate the scheduling of multiple rescue vehicles to ensure that the overall rescue efficiency is maximized without interfering with each other. When one of the rescue vehicles is unable to continue the mission due to severe road flooding or mechanical failure, the system immediately reallocates the task and hands over its target area to other nearby rescue vehicles.

[0096] S4. Real-time road environment feedback and multi-dimensional path optimization: When the rescue vehicle is on its way to the accident site, the system obtains feedback information such as road water depth and traffic congestion in real time through on-board sensors and edge computing nodes. The system combines these real-time feedback data and uses a multi-dimensional path optimization algorithm to dynamically correct the driving path of the rescue vehicle to prevent the rescue vehicle from getting stuck in traffic or being blocked by excessive water. The optimization algorithm not only considers the shortest path, but also comprehensively considers the vehicle's energy consumption and current road conditions to ensure that the rescue vehicle arrives at the accident site safely at the fastest speed.

[0097] S5. Intelligent emergency response and dynamic re-dispatch: As the weather deteriorates further, the system activates the pre-set emergency plan through the emergency response module based on reinforcement learning. The system dynamically adjusts rescue resources and transfers some rescue vehicles to secondary accident sites while maintaining the rescue force at the main accident site. Through the policy gradient algorithm, the system judges the effectiveness of the emergency strategy in real time and adjusts the strategy when necessary, such as increasing the number of rescue vehicles or changing the rescue order, to ensure that all disaster-stricken areas receive timely rescue.

[0098] S6. Dynamic collaborative learning and post-mission evaluation: After the rescue mission is completed, the system conducts a comprehensive analysis of the entire rescue process, including the rationality of path selection, the efficiency of task allocation, and resource utilization; the system automatically generates a detailed task report through a dynamic collaborative learning mechanism, and adjusts and optimizes the scheduling strategy based on feedback from actual operations to improve the rescue efficiency of similar emergencies in the future.

[0099] Application effect:

[0100] Through this second embodiment, the system has demonstrated extremely high response speed and scheduling flexibility when facing complex and changeable extreme weather and multi-vehicle chain accidents. The system not only effectively handles multi-source data fusion and real-time path planning, but can also make dynamic adjustments based on real-time feedback during the rescue process to ensure the smooth completion of the rescue mission, especially in emergency response and re-dispatching. The system can quickly initiate plans and reasonably allocate resources under extreme circumstances, significantly improving the overall rescue efficiency, reducing accident handling time, and minimizing the impact of accidents on traffic.

[0101] Implementation effect:

[0102] The method in this embodiment realizes efficient rescue dispatch in complex road environments, especially in emergency rescue under extreme weather conditions. Through dynamic data fusion and intelligent dispatching strategies, the system can significantly reduce response time, improve the success rate of rescue and overall operational efficiency, and provide strong technical support for road clearance and rescue companies.

[0103] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for automatically dispatching road obstacle removal and rescue enterprises, characterized in that: The following steps are involved: S1. Multi-source dynamic data fusion and deep environmental perception: Through the multi-source dynamic data fusion module, multi-source data such as traffic management system, meteorological information, Internet of Vehicles data, drone monitoring, smart cameras and mobile user reports are collected in real time, and dynamic convolutional neural network is used for deep environmental perception to generate a high-precision three-dimensional road environment model for predicting accidents and congestion. S2. Task priority evaluation based on multi-dimensional optimization: Combined with road conditions, weather conditions, accident severity and rescue vehicle status, a multi-dimensional optimization evaluation method based on a multi-objective genetic algorithm is used to dynamically adjust the priority of rescue tasks, and a weighted adaptive learning mechanism is used to optimize the task priority sorting strategy in real time; S3. Adaptive path planning and dynamic multi-agent scheduling: Through the multi-agent deep reinforcement learning algorithm, the system learns and optimizes path planning in real time, taking into account the real-time location of the vehicle, traffic flow, road condition changes and the urgency of the rescue mission, to achieve adaptive path planning. At the same time, the dynamic game theory algorithm is used to coordinate the coordinated scheduling of multiple rescue vehicles to ensure the overall optimality of resource allocation and path planning. S4, Real-time road environment feedback and multi-dimensional path optimization: During the execution of the system, the system obtains feedback information about the road environment in real time through on-board sensors and edge computing nodes, and uses a multi-dimensional path optimization algorithm to dynamically correct the path. The algorithm combines traffic flow prediction, vehicle energy consumption optimization, and emergency lane usage strategy to achieve optimal path planning for rescue vehicles; S5. Intelligent emergency response and dynamic re-dispatch: In the event of an emergency or extreme weather, the system automatically initiates the emergency plan through the emergency response module based on reinforcement learning, dynamically adjusts rescue resources and dispatch strategies, and ensures the priority execution of rescue tasks. The module can reallocate resources and reallocate rescue tasks in real time according to changes in road conditions to cope with multiple rescue needs. S6. Dynamic collaborative learning and post-mission evaluation: After the rescue mission is completed, the system uses a dynamic collaborative learning mechanism to analyze the path selection, task allocation and resource usage during the rescue process, generate a detailed mission report and automatically update the path planning model and scheduling strategy to achieve self-optimization and continuous improvement of the system.

2. The method for automatically dispatching road obstacle removal and rescue enterprises according to claim 1, characterized in that: In step S1, the algorithm used by the dynamic convolutional neural network is: y=f(W*X+b) Among them, y is the output feature map, W is the convolution kernel, X is the input data, b is the bias term, * is the convolution operation, and f is the activation function.

3. The automatic dispatching method for road obstacle removal and rescue enterprises according to claim 1 is characterized in that: In step S1, the parameters for establishing the three-dimensional road environment model are: Input dimension: feature dimension of multi-source data (such as image data, sensor data); Convolution kernel size: usually set to 3x3 or 5x5, depending on the data characteristics; Number of layers: The number of DCNN layers is determined by model complexity and task requirements, usually 3-10 layers; The steps of establishing the three-dimensional road environment model are: a. Data preprocessing: standardize and normalize multi-source data; b. Convolution operation: extract features from data through multi-layer convolution, and obtain feature maps at different levels layer by layer; c. 3D reconstruction: Use convolution features to generate a high-precision 3D road environment model; The multi-source data are all acquired using the latest information.

4. The automatic dispatching method for road obstacle removal and rescue enterprises according to claim 1 is characterized in that: In the step S2, the multi-objective genetic algorithm formula is: F i is the fitness value of the i-th task, w j is the weight of the j-th dimension target, dynamically adjusted, f j (x i ) is the j-th dimension objective function, x i is the feature vector of the i-th task.

5. The automatic dispatching method for road obstacle removal and rescue enterprises according to claim 1, characterized in that: In step S3, multi-agent deep reinforcement learning is combined with the dynamic game theory algorithm, and the corresponding algorithm formula is: Among them, Q(s,a) is the value of taking action a in state s, α is the learning rate, r is the immediate reward, γ is the discount factor, weighing future rewards, s′ is the next state, is the maximum expected return in the next state; When the intelligent agent dispatching system detects a failure of a rescue vehicle, it can automatically reallocate tasks and adjust the paths of other vehicles.

6. The method for automatically dispatching road obstacle removal and rescue enterprises according to claim 1, characterized in that: In the step S4, the multi-dimensional path optimization algorithm formula is: Among them, L total is the total path cost, w k is the weight of the k-th dimension path factor, L k is the k-th dimension path cost; The dynamic correction is performed sequentially according to the urgency and reliability of the feedback.

7. The automatic dispatching method for road obstacle removal and rescue enterprises according to claim 1, characterized in that: In the step S5, the dynamic adjustment is determined according to the policy gradient algorithm, and the formula of the policy gradient algorithm is: Among them, J(θ) is the expected return of the strategy, π θ (a|s) is the probability of taking action a in state s, is the value function; The dispatching resources will reserve a certain proportion of rescue resources to deal with emergencies.

8. The method for automatically dispatching road obstacle removal and rescue enterprises according to claim 1, characterized in that: In step S6, the formula used by the dynamic collaborative learning mechanism is: Among them, Δθ is the update value of the model parameters, η is the learning rate, (y i ,f(x i ; θ)) is the task loss function, x i and i is the data and results of the i-th task.

Images