Intelligent vehicle scheduling system and method fusing ai multi-dimensional perception and dynamic optimization

Abstract

This invention relates to the field of intelligent scheduling technology, and discloses an intelligent vehicle scheduling system and method integrating AI multi-dimensional perception and dynamic optimization. The method includes the following steps: S1: acquiring the starting point location information, ending point location information, and real-time operating data of multiple available vehicles for multiple transportation tasks within a target area; S2: extracting features from the real-time operating data, starting point location information, and ending point location information to generate multi-dimensional scheduling feature information. This invention uses a feature extraction module to perform deep processing on the vehicle's current location information, vehicle load status information, vehicle operating status information, and the starting point and ending point location information of transportation tasks, generating multi-dimensional scheduling feature information that includes spatial distance features, load matching features, and available time features. This achieves comprehensive perception and quantitative expression of scheduling elements, providing a rich data foundation for subsequent scheduling decisions.

Description

Technical Field

[0001] This invention relates to the field of intelligent scheduling technology, specifically to an intelligent vehicle scheduling system and method that integrates AI multi-dimensional perception and dynamic optimization. Background Technology

[0002] With the rapid development of e-commerce and logistics, the number of urban delivery vehicles has surged. Efficiently dispatching vehicles to complete a large volume of transportation tasks has become a significant challenge for logistics companies. Traditional vehicle dispatching methods rely heavily on dispatchers' manual experience, allocating vehicles manually based on task location and vehicle load capacity. When dealing with hundreds of vehicles and thousands of tasks, dispatchers spend considerable time on task matching and route planning, resulting in inefficient scheduling and difficulty in ensuring global optimization. More importantly, traditional dispatching methods cannot dynamically adjust routes based on real-time traffic conditions after vehicle departure. When encountering traffic congestion or emergencies, vehicles often have to continue along their original routes or detour under the guidance of individual drivers, leading to transportation delays and increased energy consumption. While some static model-based dispatching systems have emerged, these systems typically only consider the basic matching relationship between tasks and vehicles, lacking multi-dimensional perception of real-time vehicle operating status and failing to integrate real-time traffic information into the route optimization process. They are ill-suited to the complex and ever-changing urban traffic environment, resulting in limited improvements in transportation efficiency and failing to meet the actual needs of logistics companies for intelligent and automated dispatching. Summary of the Invention

[0003] The purpose of this invention is to provide an intelligent vehicle dispatching system and method that integrates AI multi-dimensional perception and dynamic optimization to solve the problems mentioned in the background art.

[0004] To achieve the above objectives, the present invention provides the following technical solution: an intelligent vehicle scheduling method integrating AI multi-dimensional perception and dynamic optimization, comprising the following steps: S1: Obtain the starting point location information, ending point location information, and real-time operation data of multiple available vehicles for multiple transportation tasks within the target area; S2: Extract features from real-time running data, as well as starting and ending point location information, to generate multi-dimensional scheduling feature information; S3: Input the multi-dimensional scheduling feature information into the pre-trained scheduling decision model and output a preliminary scheduling plan for multiple transportation tasks; wherein, the scheduling decision model is constructed based on a machine learning model and is used to represent the mapping relationship between multi-dimensional scheduling feature information and vehicle scheduling decisions; S4: Obtain real-time traffic information for the target area, and perform dynamic path optimization on the preliminary scheduling plan based on the real-time traffic information to generate the optimal driving path for each available vehicle. S5: Generates scheduling instructions based on the optimal driving route and sends the scheduling instructions to the corresponding available vehicles, controlling the available vehicles to perform transportation tasks.

[0005] As a preferred embodiment of the present invention, the real-time operation data includes vehicle current location information, vehicle load status information, and vehicle operation status information; the step S2, which extracts features from the real-time operation data, starting point location information, and ending point location information to generate multi-dimensional scheduling feature information, includes: Spatial distance analysis is performed between the vehicle's current location information and its starting point location information to obtain the first feature data; The steps for performing a matching degree analysis between vehicle load status information and cargo information of transportation tasks to obtain second feature data include: comparing the vehicle's rated load with the cargo weight of the transportation task; when the vehicle's remaining load is greater than the cargo weight, calculating the normalized score of the difference between the two as the second feature data. The available time of a vehicle is determined based on its operating status information, and the first feature data, the second feature data, and the available time are integrated to generate multi-dimensional scheduling feature information.

[0006] As a preferred embodiment of the present invention, the step in S4 of dynamically optimizing the preliminary scheduling plan based on real-time traffic information to generate the optimal driving path for each available vehicle includes: The task sequence and corresponding waypoint information for each available vehicle are determined based on the preliminary scheduling plan. Obtain real-time traffic flow and traffic incident information within the target area as real-time traffic information; The information on waypoints is integrated with real-time traffic information to generate a dynamic road network model. The steps for generating the dynamic road network model include: dynamically adjusting the road weight values ​​of each road segment in the dynamic road network model based on real-time traffic flow information, wherein the road weight value of congested road segments is increased and the road weight value of unobstructed road segments is decreased. The driving paths in the task sequence are iteratively optimized based on the dynamic road network model to generate the optimal driving path for each available vehicle to avoid congested road sections.

[0007] As a preferred embodiment of the present invention, before the step of obtaining the starting location information, ending location information, and real-time operation data of multiple available vehicles for multiple transportation tasks within the target area in step S1, the method further includes: Collect historical transportation task data and corresponding historical scheduling scheme data to construct a training sample set; Using historical transportation task data from the training sample set as input and historical scheduling scheme data as supervision labels, the initial neural network model is trained to obtain the scheduling decision model.

[0008] As a preferred embodiment of the present invention, after the step of generating a scheduling instruction based on the optimal driving route and sending the scheduling instruction to the corresponding available vehicles in step S5, and controlling the available vehicles to perform the transportation task, the method further includes: Real-time monitoring of the actual driving trajectory of available vehicles during the transportation task; When the actual driving trajectory is detected to deviate from the optimal driving path and the deviation distance exceeds a preset threshold, a path replanning instruction is triggered. Based on the current location of available vehicles and real-time traffic information, an updated optimal driving route is regenerated. Preferably, the preset threshold can be set according to the actual road network conditions, for example, it can be set to 50 meters or 100 meters.

[0009] An intelligent vehicle dispatching system integrating AI multi-dimensional perception and dynamic optimization, used to implement any one of the methods described above, includes: The data acquisition module is used to acquire the starting point location information, ending point location information, and real-time operation data of multiple available vehicles for multiple transportation tasks within the target area; The feature extraction module is used to extract features from real-time running data, as well as start-up and end-up location information, to generate multi-dimensional scheduling feature information. The scheme generation module is used to input multi-dimensional scheduling feature information into a pre-trained scheduling decision model and output preliminary scheduling schemes for multiple transportation tasks; wherein, the scheduling decision model is constructed based on a machine learning model and is used to represent the mapping relationship between multi-dimensional scheduling feature information and vehicle scheduling decisions; The route optimization module is used to obtain real-time traffic information of the target area and perform dynamic route optimization on the preliminary scheduling plan based on the real-time traffic information to generate the optimal driving route for each available vehicle. The instruction execution module is used to generate scheduling instructions based on the optimal driving route and send the scheduling instructions to the corresponding available vehicles to control the available vehicles to perform transportation tasks.

[0010] As a preferred embodiment of the present invention, the real-time operating data includes vehicle current location information, vehicle load status information, and vehicle operating status information; the feature extraction module includes: The first analysis unit is used to perform spatial distance analysis on the vehicle's current location information and the starting point location information to obtain the first feature data; The second analysis unit is used to perform a matching degree analysis on the vehicle load status information and the cargo information of the transportation task to obtain the second feature data. The second feature data includes: comparing the vehicle's rated load with the cargo weight of the transportation task. When the vehicle's remaining load is greater than the cargo weight, the normalized score of the difference between the two is calculated as the second feature data. The data integration unit is used to determine the available time of a vehicle based on the vehicle's operating status information, and to integrate the first feature data, the second feature data, and the available time to generate multi-dimensional scheduling feature information.

[0011] As a preferred embodiment of the present invention, the path optimization module includes: The task parsing unit is used to determine the task sequence and corresponding waypoint information for each available vehicle based on the preliminary scheduling plan. The traffic data acquisition unit is used to acquire real-time traffic flow information and traffic event information within the target area, as real-time traffic information. The model fusion unit is used to fuse waypoint information with real-time traffic information to generate a dynamic road network model. The generation of the dynamic road network model includes: dynamically adjusting the road weight values ​​of each road segment in the dynamic road network model according to real-time traffic flow information, wherein the road weight value of congested road segments is increased and the road weight value of unobstructed road segments is decreased. The path iteration unit is used to iteratively optimize the driving paths in the task sequence based on the dynamic road network model, and generate the optimal driving path for each available vehicle to avoid congested road sections.

[0012] As a preferred technical solution of the present invention, the system further includes a model training module, which is used to collect historical transportation task data and corresponding historical scheduling scheme data to construct a training sample set before the data acquisition module acquires the starting location information, ending location information and real-time operation data of multiple transportation tasks in the target area and multiple available vehicles. Using historical transportation task data from the training sample set as input and historical scheduling scheme data as supervision labels, the initial neural network model is trained to obtain the scheduling decision model.

[0013] As a preferred technical solution of the present invention, the system further includes a dynamic adjustment module, which is used to monitor the actual driving trajectory of the available vehicles in real time after the instruction execution module generates a scheduling instruction based on the optimal driving path, sends the scheduling instruction to the corresponding available vehicles, and controls the available vehicles to perform the transportation task. When the actual driving trajectory is detected to deviate from the optimal driving path and the deviation distance exceeds a preset threshold, a path replanning instruction is triggered; based on the current location of available vehicles and real-time traffic information, an updated optimal driving path is regenerated.

[0014] Compared with the prior art, the beneficial effects of the present invention are: 1. This invention uses a feature extraction module to perform in-depth processing on the vehicle's current location information, vehicle load status information, vehicle operation status information, and the starting and ending location information of the transportation task, generating multi-dimensional scheduling feature information that includes spatial distance features, load matching features, and available time features. This enables comprehensive perception and quantitative expression of scheduling elements, providing a rich data foundation for subsequent scheduling decisions.

[0015] 2. This invention processes multi-dimensional scheduling feature information through a scheduling decision model built on a machine learning model, and outputs preliminary scheduling schemes for multiple transportation tasks. This model can automatically learn the inherent rules and optimization strategies of task allocation from historical scheduling data, overcoming the shortcomings of traditional manual scheduling that relies on subjective experience and is difficult to guarantee global optimality, and significantly improving the efficiency and scientific nature of scheduling scheme formulation.

[0016] 3. This invention obtains real-time traffic information of the target area through a path optimization module, and performs dynamic path optimization processing on the preliminary scheduling plan based on the real-time traffic information to generate the optimal driving path for each available vehicle. This process integrates real-time traffic flow information and traffic event information into the path planning, enabling vehicles to effectively avoid congested road sections, significantly shorten travel time and mileage, and reduce transportation energy consumption costs.

[0017] 4. This invention generates dispatch instructions based on the optimal driving route through the instruction execution module and automatically sends them to the corresponding vehicles, realizing an automated closed loop from dispatch decision to vehicle execution, reducing manual intervention and improving the timeliness and accuracy of dispatch response.

[0018] 5. This invention uses a dynamic adjustment module to monitor the actual driving trajectory of the vehicle in real time during the execution of the task. When the vehicle is detected to deviate from the optimal driving path and exceed a preset threshold, a path replanning instruction is automatically triggered. The updated optimal driving path is regenerated based on the vehicle's current location and real-time traffic information, enabling the scheduling system to have dynamic adaptive capabilities to cope with emergencies, effectively reducing the task timeout rate and emergency response time. Attached Figure Description

[0019] Figure 1 This is a schematic diagram of the overall process of the intelligent vehicle scheduling method that integrates AI multi-dimensional perception and dynamic optimization according to the present invention. Figure 2 This is a schematic diagram of the modular structure of the intelligent vehicle dispatching system that integrates AI multi-dimensional perception and dynamic optimization according to the present invention. Detailed Implementation

[0020] The technical solutions of the present invention will be clearly and completely described below with reference to the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0021] Example 1

[0022] An intelligent vehicle scheduling method integrating AI multi-dimensional perception and dynamic optimization includes the following steps: S1: Obtain the starting point location information, ending point location information, and real-time operation data of multiple available vehicles for multiple transportation tasks within the target area; S2: Extract features from real-time running data, as well as starting and ending point location information, to generate multi-dimensional scheduling feature information; S3: Input the multi-dimensional scheduling feature information into the pre-trained scheduling decision model and output a preliminary scheduling plan for multiple transportation tasks; wherein, the scheduling decision model is built based on a machine learning model and is used to represent the mapping relationship between multi-dimensional scheduling feature information and vehicle scheduling decisions; S4: Obtain real-time traffic information for the target area, and perform dynamic path optimization on the preliminary scheduling plan based on the real-time traffic information to generate the optimal driving path for each available vehicle. S5: Generates scheduling instructions based on the optimal driving route and sends the scheduling instructions to the corresponding available vehicles, controlling the available vehicles to perform transportation tasks.

[0023] Furthermore, the real-time operational data includes the vehicle's current location information, vehicle load status information, and vehicle operational status information; the steps in S2 to extract features from the real-time operational data, as well as the starting point and ending point location information, to generate multi-dimensional scheduling feature information include: Spatial distance analysis is performed between the vehicle's current location information and its starting point location information to obtain the first feature data; The steps for performing a matching degree analysis between vehicle load status information and cargo information of transportation tasks to obtain second feature data include: comparing the vehicle's rated load with the cargo weight of the transportation task; when the vehicle's remaining load is greater than the cargo weight, calculating the normalized score of the difference between the two as the second feature data. The available time of a vehicle is determined based on its operating status information, and the first feature data, the second feature data, and the available time are integrated to generate multi-dimensional scheduling feature information.

[0024] Furthermore, in S4, the step of dynamically optimizing the initial scheduling plan based on real-time traffic information to generate the optimal travel route for each available vehicle includes: The task sequence and corresponding waypoint information for each available vehicle are determined based on the preliminary scheduling plan. Obtain real-time traffic flow and traffic incident information within the target area as real-time traffic information; The information on waypoints is integrated with real-time traffic information to generate a dynamic road network model. The steps for generating the dynamic road network model include: dynamically adjusting the road weight values ​​of each road segment in the dynamic road network model based on real-time traffic flow information, wherein the road weight value of congested road segments is increased and the road weight value of unobstructed road segments is decreased. The driving paths in the task sequence are iteratively optimized based on the dynamic road network model to generate the optimal driving path for each available vehicle to avoid congested road sections.

[0025] Furthermore, before the step in S1 of obtaining the start-up location information, end-point location information, and real-time operation data of multiple available vehicles for multiple transportation tasks within the target area, the following steps are also included: Collect historical transportation task data and corresponding historical scheduling scheme data to construct a training sample set; Using historical transportation task data from the training sample set as input and historical scheduling scheme data as supervision labels, the initial neural network model is trained to obtain the scheduling decision model.

[0026] Furthermore, after the steps in S5 of generating dispatch instructions based on the optimal driving route, sending the dispatch instructions to the corresponding available vehicles, and controlling the available vehicles to execute the transportation task, it also includes: Real-time monitoring of the actual driving trajectory of available vehicles during the transportation task; When the actual driving trajectory is detected to deviate from the optimal driving path and the deviation distance exceeds a preset threshold, a path replanning instruction is triggered. Based on the current location of available vehicles and real-time traffic information, an updated optimal driving route is regenerated. Preferably, the preset threshold can be set according to the actual road network conditions, for example, it can be set to 50 meters or 100 meters.

[0027] An intelligent vehicle dispatching system integrating AI multi-dimensional perception and dynamic optimization, comprising methods for achieving any of the above, including: The data acquisition module is used to acquire the starting point location information, ending point location information, and real-time operation data of multiple available vehicles for multiple transportation tasks within the target area; The feature extraction module is used to extract features from real-time running data, as well as start-up and end-up location information, to generate multi-dimensional scheduling feature information. The scheme generation module is used to input multi-dimensional scheduling feature information into a pre-trained scheduling decision model and output preliminary scheduling schemes for multiple transportation tasks. The scheduling decision model is built based on a machine learning model and is used to represent the mapping relationship between multi-dimensional scheduling feature information and vehicle scheduling decisions. The route optimization module is used to obtain real-time traffic information of the target area and perform dynamic route optimization on the preliminary scheduling plan based on the real-time traffic information to generate the optimal driving route for each available vehicle. The instruction execution module is used to generate scheduling instructions based on the optimal driving route and send the scheduling instructions to the corresponding available vehicles to control the available vehicles to perform transportation tasks.

[0028] Furthermore, real-time operational data includes the vehicle's current location information, vehicle load status information, and vehicle operational status information; the feature extraction module includes: The first analysis unit is used to perform spatial distance analysis on the vehicle's current location information and the starting point location information to obtain the first feature data; The second analysis unit is used to perform a matching degree analysis on the vehicle load status information and the cargo information of the transportation task to obtain the second feature data. The second feature data includes: comparing the vehicle's rated load with the cargo weight of the transportation task. When the vehicle's remaining load is greater than the cargo weight, the normalized score of the difference between the two is calculated as the second feature data. The data integration unit is used to determine the available time of a vehicle based on the vehicle's operating status information, and to integrate the first feature data, the second feature data, and the available time to generate multi-dimensional scheduling feature information.

[0029] Furthermore, the path optimization module includes: The task parsing unit is used to determine the task sequence and corresponding waypoint information for each available vehicle based on the preliminary scheduling plan. The traffic data acquisition unit is used to acquire real-time traffic flow information and traffic event information within the target area, as real-time traffic information. The model fusion unit is used to fuse waypoint information with real-time traffic information to generate a dynamic road network model. The generation of the dynamic road network model includes: dynamically adjusting the road weight values ​​of each road segment in the dynamic road network model according to real-time traffic flow information, wherein the road weight value of congested road segments is increased and the road weight value of unobstructed road segments is decreased. The path iteration unit is used to iteratively optimize the driving paths in the task sequence based on the dynamic road network model, and generate the optimal driving path for each available vehicle to avoid congested road sections.

[0030] Furthermore, the system also includes a model training module, which is used to collect historical transportation task data and corresponding historical scheduling scheme data to build a training sample set before the data acquisition module acquires the starting location information, ending location information and real-time operation data of multiple transportation tasks in the target area and multiple available vehicles. Using historical transportation task data from the training sample set as input and historical scheduling scheme data as supervision labels, the initial neural network model is trained to obtain the scheduling decision model.

[0031] Furthermore, the system also includes a dynamic adjustment module, which is used to monitor the actual driving trajectory of available vehicles in real time after the instruction execution module generates scheduling instructions based on the optimal driving route, sends the scheduling instructions to the corresponding available vehicles, and controls the available vehicles to perform transportation tasks. When the actual driving trajectory is detected to deviate from the optimal driving path and the deviation distance exceeds a preset threshold, a path replanning instruction is triggered; based on the current location of available vehicles and real-time traffic information, an updated optimal driving path is regenerated.

[0032] Example 2

[0033] This embodiment uses a logistics distribution center in a first-tier city as an application scenario to specifically verify the intelligent vehicle scheduling method that integrates AI multi-dimensional perception and dynamic optimization. The logistics distribution center handles approximately 8,000 transportation tasks per day and has more than 300 delivery vehicles of various types, with its service area covering the main urban area and surrounding suburbs of the city.

[0034] In the specific implementation process, the data acquisition module first collects the transportation task information to be executed in the target area that day, including the starting location information of each task, i.e. the loading point of the goods, and the ending location information, i.e. the customer's receiving point. At the same time, it acquires the real-time operation data of all available vehicles. The real-time operation data specifically includes the current location information of the vehicle obtained through the vehicle-mounted GPS terminal, the vehicle load status information read by the vehicle-mounted sensors, the current load weight, and the vehicle operation status information, including vehicle speed, remaining battery or fuel level, and vehicle health status.

[0035] The feature extraction module processes the above data. When performing spatial distance analysis between the vehicle's current location information and the starting point location information, it uses Gaussian projection to convert latitude and longitude coordinates into planar coordinates and calculates the Euclidean distance as the first feature data. When performing matching degree analysis between the vehicle's load status information and the cargo information of the transportation task, it compares the vehicle's rated load with the weight of the task cargo. When the vehicle's remaining load is greater than the weight of the task cargo and the difference is within a reasonable range, it calculates the normalized score of the difference between the two and assigns a higher matching degree score as the second feature data. Based on the remaining battery or fuel and average energy consumption data in the vehicle's operating status information, it estimates the available time for the vehicle to operate continuously. After normalizing the above first feature data, second feature data, and available time, it integrates them to generate multi-dimensional scheduling feature information.

[0036] The scheme generation module inputs multi-dimensional scheduling feature information into a pre-trained scheduling decision model. This scheduling decision model is constructed using a gradient boosting decision tree framework. The training data comes from the logistics distribution center's historical transportation task data and corresponding historical scheduling scheme data over the past six months. The model outputs a preliminary scheduling scheme for all transportation tasks on the day, including a list of tasks that each vehicle needs to perform and the execution order of the tasks.

[0037] The route optimization module acquires real-time traffic information for the city, including real-time traffic flow data from the traffic management department interface and information on traffic events such as traffic accidents and road construction. Based on the preliminary scheduling plan, it determines the task sequence and corresponding waypoint information for each vehicle. The waypoint information is then fused with the real-time traffic information to construct a dynamic road network model. In the dynamic road network model, the road weights of each road segment are dynamically adjusted based on the real-time traffic flow information. The road weights of congested road segments are increased, while the road weights of uncongested road segments are decreased. Based on the dynamic road network model, the ant colony optimization algorithm is used to iteratively optimize the driving paths in the task sequence. After 200 iterations, the algorithm converges, generating the optimal driving path for each vehicle to avoid congested road segments.

[0038] The instruction execution module generates dispatch instructions based on the optimal driving route and sends the dispatch instructions to the vehicle's onboard terminal via a 4G / 5G wireless communication network. The onboard terminal guides the driver to perform the transportation task through voice prompts and navigation interface display.

[0039] Example 3

[0040] This embodiment takes the raw material transportation scheduling of a manufacturing enterprise as an application scenario to further verify the applicability of the technical solution of the present invention in the field of industrial logistics. The manufacturing enterprise has three factory areas located in the east, west and north of the city, and needs to transport production materials from five raw material suppliers on a regular basis. The daily transportation task is about 200 truck trips, with a total of 80 vehicles, including fuel trucks and electric trucks.

[0041] In the specific implementation process, the data acquisition module collects the material demand tasks of each factory area. Each task includes the starting location information, i.e., the location of the supplier's warehouse, and the ending location information, i.e., the location of the target factory area. At the same time, it acquires the real-time operation data of the vehicle. It is worth noting that in this embodiment, the remaining battery power information of the electric truck is additionally collected as an important part of the vehicle's operating status information.

[0042] When the feature extraction module performs feature extraction, it uses Manhattan distance instead of Euclidean distance to analyze the spatial distance between the vehicle's current location information and the starting point location information, in order to adapt to the characteristics of the grid layout of urban roads. When performing matching degree analysis between the vehicle's load status information and the cargo information of the transportation task, in addition to considering load matching, it also considers the compatibility between cargo type and vehicle body type. For example, some materials require closed body transportation, while some materials require flatbed truck transportation. These factors are included in the matching degree analysis to form the second feature data. When determining the vehicle's available time based on the vehicle's operating status information, it estimates the remaining fuel and average fuel consumption for fuel vehicles, and estimates the remaining power and average power consumption for electric vehicles. The charging demand of electric vehicles is also taken into consideration. When the remaining power is lower than a set threshold, the available time is reduced by the charging time accordingly.

[0043] The scheme generation module inputs multi-dimensional scheduling feature information into the scheduling decision model. The scheduling decision model in this embodiment is constructed using a deep neural network architecture, which includes an input layer, three hidden layers and an output layer. The number of nodes in the hidden layers are 128, 64 and 32 respectively. The model training data comes from the transportation scheduling records of the manufacturing company over the past year, totaling approximately 73,000 sample data.

[0044] When the route optimization module obtains real-time traffic information, in addition to accessing publicly available data from traffic management departments, it also uses GPS trajectory data of the company's own vehicles to infer the traffic speed of each road segment as a supplement to the real-time traffic information. After fusing waypoint information with real-time traffic information to generate a dynamic road network model, the module dynamically adjusts the road weight values ​​of each road segment in the dynamic road network model based on real-time traffic flow information. Specifically, the road weight values ​​of congested road segments are increased, while the road weight values ​​of unobstructed road segments are decreased. A genetic algorithm is used for iterative route optimization, with a population size of 100, a crossover probability of 0.8, and a mutation probability of 0.1. After 150 generations of evolution, the algorithm converges to generate the optimal driving path for each vehicle.

[0045] The instruction execution module converts the generated optimal driving route into a dispatch instruction and sends it to the vehicle terminal. For electric vehicles, the dispatch instruction specifically marks the location of charging facilities along the route so that the vehicle can be recharged in time when the battery is low.

[0046] Compare with Example 1 This comparative example uses the traditional static vehicle scheduling method for comparison and verification. The application scenario is the same as that in Example 2, which is the daily transportation scheduling of a logistics distribution center in a first-tier city.

[0047] In the specific implementation process, this comparative example does not include the multi-dimensional perception feature extraction step, nor does it include the scheduling decision model based on machine learning. The dispatcher manually divides the transportation tasks according to geographical location based on experience, usually with administrative divisions or street boundaries as reference, and assigns tasks in the same area to the same vehicle. The vehicle allocation mainly considers the matching of the vehicle's load capacity with the total weight of the task, without considering the spatial distance relationship between the vehicle's current location and the task's starting point, or the vehicle's operating status information.

[0048] The route planning adopts the static shortest path method, which pre-calculates the shortest distance path between each task point based on the city's electronic map. It does not consider changes in real-time traffic information. After the vehicle departs, it travels along the pre-planned fixed route. Even if it encounters traffic congestion, it continues to move along the original route without making dynamic adjustments.

[0049] When a vehicle deviates from the planned route during a mission, the dispatcher communicates with the driver by phone to understand the situation and manually determines whether subsequent tasks need to be adjusted. This method cannot automatically trigger route replanning, and the entire dispatching process relies on human experience and manual operation. Dispatch instructions are transmitted via telephone or walkie-talkie, lacking an automated instruction generation and issuance mechanism.

[0050] The formulation of the dispatching plan in this example takes a long time, usually requiring 2 to 3 hours to complete the initial allocation of all tasks for the day. During the dispatching process, uneven task allocation is prone to occur, with some vehicles being overloaded and working overtime, while others are underloaded and waste capacity. When faced with sudden traffic congestion or vehicle breakdowns, the response speed is slow, and the adjustment plan often lags behind changes in the actual situation.

[0051] Compare with Example 2 This comparative example uses a method that includes only a machine learning scheduling decision model but not dynamic path optimization for comparison and verification. The application scenario is the same as that in Example 3, which is the raw material transportation scheduling of a manufacturing company.

[0052] In the specific implementation process, this comparative example includes data acquisition steps and feature extraction steps, which are basically the same as those in Example 3. The feature extraction module also processes the vehicle's current location information, vehicle load status information, and vehicle operation status information to generate multi-dimensional scheduling feature information.

[0053] The scheme generation module also inputs multi-dimensional scheduling feature information into the pre-trained scheduling decision model and outputs a preliminary scheduling scheme for the transportation task. The structure and training method of the scheduling decision model are consistent with those in Example 3, ensuring that the only control variable is whether or not the dynamic path optimization step is included.

[0054] However, this comparative example does not include the step of obtaining real-time traffic information, nor does it include the step of dynamically optimizing the preliminary scheduling plan based on real-time traffic information. After generating the preliminary scheduling plan, the fixed path is directly calculated according to the task sequence determined in the plan. The path calculation adopts a static road network model, and the road weight values ​​of each road segment remain fixed and are set based on the historical average travel time, without being adjusted according to real-time traffic conditions.

[0055] The instruction execution module generates scheduling instructions based on a fixed path and sends them to the vehicle terminal. The vehicle travels along the fixed path to perform the transportation task. When encountering traffic congestion along the way, it cannot obtain dynamic detour guidance and can only rely on the driver's familiarity with the local road conditions to judge whether to detour. The difference in judgment ability among different drivers leads to large fluctuations in travel time and fuel consumption.

[0056] When the actual driving trajectory of a vehicle deviates from the fixed path, the system does not trigger path replanning, and subsequent tasks are still executed according to the original plan. If the deviation causes the task sequence to be unable to continue, manual intervention by the dispatcher is required. The lack of an automated adjustment mechanism means that although this comparative example has achieved preliminary scheduling automation through a machine learning model, the lack of dynamic response capability to real-time road conditions significantly restricts the improvement of transportation efficiency in actual execution.

[0057] Experimental Data and Effect Analysis To verify the superior effect of the technical solution of the present invention, a comparative experiment was conducted for three months in the application scenarios of the above four embodiments and control examples. Detailed operating data was collected and statistically analyzed. During the experiment, each embodiment and control example operated normally in its respective application scenario, and key indicators such as daily transportation task completion, vehicle running time, total mileage, fuel or electricity consumption were recorded. The experimental data is summarized in Table 1.

[0058] Table 1 Comparison of three-month operating data between each embodiment and the control example A thorough analysis of the data in the table above reveals that Examples 1, 2, and 3 are significantly superior to Comparative Examples 1 and 2 in all key indicators, fully demonstrating the advanced nature and practicality of the technical solution of the present invention. Among them, Example 1, as the basic implementation scheme, has shown obvious advantages in all indicators; Example 2 is further optimized in the logistics and distribution scenario of first-tier cities, with some indicators slightly improved compared to Example 1; Example 3 also verifies the applicability of the technical solution in the industrial logistics scenario.

[0059] In terms of average daily task completion, Example 1 reached 8150 orders, an increase of 6.9% compared to 7624 orders in Control Example 1, and an increase of 8.5% compared to 189 orders in Control Example 2 with the same vehicle scale. Example 2 reached 8236 orders, an increase of 8.0% compared to Control Example 1 and 9.5% compared to Control Example 2. This data shows that the technical solution of the present invention can make fuller use of existing transportation resources and complete more transportation tasks with the same number of vehicles. The increase in the average daily task completion per vehicle is more intuitive. Example 1 completed 27.2 orders per vehicle per day, an increase of 7.1% compared to 25.4 orders in Control Example 1. Example 2 completed 27.5 orders per vehicle per day, an increase of 8.3% compared to Control Example 1. Example 3 completed 2.6 orders per vehicle per day, an increase of 8.3% compared to 2.4 orders in Control Example 2. The data of each group corroborate each other, indicating that the improvement effect is stable and repeatable.

[0060] The average task response time refers to the time interval from task generation to vehicle receiving dispatch instructions. The average task response time in Example 1 is 8.5 minutes, and in Example 2 it is 8.3 minutes, both significantly lower than the 15.6 minutes in Comparative Example 1 and the 14.2 minutes in Comparative Example 2. This advantage mainly stems from the efficiency of the automated dispatch process of this invention. Comparative Example 1 relied on manual dispatch, which took as long as 156 minutes to complete the formulation of the daily dispatch plan, resulting in a significant extension of the task response time. Although Comparative Example 2 automatically generated the dispatch plan in a shorter time, due to the lack of dynamic path optimization, adjustments during subsequent execution still required manual intervention, affecting the overall response efficiency.

[0061] Average single-mission mileage and average single-mission time are core indicators for measuring transportation efficiency. In Example 1, the average single-mission mileage was 6.9 kilometers, which was 15.9% shorter than that of Control Example 1 (8.2 kilometers) and 67.9% shorter than that of Control Example 2 (21.5 kilometers). The average single-mission time was 19.2 minutes, which was 28.4% shorter than that of Control Example 1 (26.8 minutes) and 56.1% shorter than that of Control Example 2 (43.7 minutes). In Example 2, the average single-mission mileage was 6.8 kilometers, which was 17.1% shorter than that of Control Example 1 and 68.4% shorter than that of Control Example 2. The average single-task travel time was 18.5 minutes, which was 31.0% shorter than that of Control Example 1 and 57.7% shorter than that of Control Example 2. The average single-task travel distance of Example 3 was 18.4 kilometers, which was 14.4% shorter than that of Control Example 2 (21.5 kilometers); the average single-task travel time was 31.2 minutes, which was 28.6% shorter than that of Control Example 2 (43.7 minutes). These data fully demonstrate that the present invention effectively reduces the empty driving mileage and congestion waiting time of vehicles through the synergistic effect of multi-dimensional perception feature extraction, machine learning scheduling decision-making, and dynamic path optimization, thereby achieving the optimization of transportation routes.

[0062] The average energy cost per task directly reflects the economic benefits. The average energy cost per task in Example 1 is 4.3 yuan, which is 24.6% lower than the 5.7 yuan in Control Example 1 and 70.9% lower than the 14.8 yuan in Control Example 2. Based on an average of 8,150 tasks per day, Example 1 can save approximately 11,410 yuan in energy costs per day, accumulating to over 1.02 million yuan in savings over three months. The average energy cost per task in Example 2 is 4.2 yuan, which is 26.3% lower than the 1 in Control Example 1 and 71.6% lower than the 2 in Control Example 2. Based on an average of 8,236 tasks per day, it saves approximately 12,354 yuan in energy costs per day, accumulating to over 1.11 million yuan in savings over three months. The average energy cost per task in Example 3 is 11.6 yuan, which is 21.6% lower than the 14.8 yuan in Control Example 2. It saves approximately 662 yuan in energy costs per day, accumulating to approximately 60,000 yuan in savings over three months. These data strongly demonstrate the outstanding effect of the technical solution of the present invention in energy conservation and emission reduction, and the economic benefits are very significant.

[0063] Regarding the time required to generate scheduling schemes, Examples 1, 2, and 3 only require 1.9 minutes, 1.8 minutes, and 0.9 minutes, respectively, to complete the scheduling decisions for all tasks. In contrast, Example 1 requires 156 minutes, or 2.6 hours, a difference of nearly two orders of magnitude. This huge gap fully demonstrates the efficiency advantage of machine learning-based automated scheduling compared to traditional manual scheduling. It frees schedulers from tedious daily allocation work, allowing them to focus on handling abnormal situations and optimizing overall operational strategies. Example 3 has the shortest time consumption, which is related to its smaller workload, but its automation efficiency advantage is still obvious.

[0064] Task timeout rate is an important indicator for measuring the effectiveness of scheduling schemes. The task timeout rate of Example 1 was 3.5%, Example 2 was 3.2%, and Example 3 was 5.1%, which were significantly lower than the 12.8% of Control Example 1 and the 9.6% of Control Example 2. This result shows that the technical solution of the present invention can not only formulate a reasonable scheduling plan, but more importantly, through dynamic path optimization and real-time adjustment mechanisms, ensure that the plan can effectively cope with various emergencies during actual execution, thereby ensuring that the task is completed on time. Although Control Example 2 adopted a machine learning scheduling model, it lacked dynamic path optimization and had insufficient ability to cope with real-time changes in traffic conditions, resulting in a significantly higher task timeout rate than Example 1 and Example 2.

[0065] Vehicle idle rate reflects the utilization efficiency of transportation capacity resources. The vehicle idle rate of Example 1 is 5.1%, Example 2 is 4.6%, and Example 3 is 3.9%, which is much lower than the 12.3% of Comparative Example 1 and 8.2% of Comparative Example 2. This data shows that the technical solution of the present invention can allocate transportation tasks more evenly, avoiding the situation where some vehicles are overworked while others are idle and wasted, and achieving the optimal allocation of transportation capacity resources. Comparative Example 1 relies on manual experience for allocation, which is prone to uneven task allocation due to subjective judgment bias. Although Comparative Example 2 optimizes the initial allocation through a machine learning model, the lack of dynamic adjustment during the execution process means that the balance of subsequent task allocation will also be affected when changes in actual road conditions cause delays for some vehicles.

[0066] Emergency response time measures a system's ability to react quickly to unexpected situations. Example 1's emergency response time is 1.6 minutes, Example 2's is 1.5 minutes, and Example 3's is 1.8 minutes, significantly better than Comparative Example 1's 18.4 minutes and Comparative Example 2's 16.7 minutes. This advantage stems from the real-time monitoring and automatic replanning mechanism in this invention. When a vehicle deviates from its predetermined path or encounters an emergency, the system can immediately trigger a path replanning command, generating an updated optimal driving path in a very short time. Comparative Example 1 requires a dispatcher to manually understand the situation, analyze the problem, formulate a plan, and transmit instructions, a process that is time-consuming. Although Comparative Example 2 can use a machine learning model to assist human decision-making, it still requires human intervention, resulting in a significantly insufficient response speed.

[0067] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention.

Claims

1. An intelligent vehicle scheduling method integrating AI multi-dimensional perception and dynamic optimization, characterized in that, Includes the following steps: S1: Obtain the starting point location information, ending point location information, and real-time operation data of multiple available vehicles for multiple transportation tasks within the target area; S2: Extract features from real-time running data, as well as starting and ending point location information, to generate multi-dimensional scheduling feature information; S3: Input the multi-dimensional scheduling feature information into the pre-trained scheduling decision model and output a preliminary scheduling plan for multiple transportation tasks; wherein, the scheduling decision model is constructed based on a machine learning model and is used to represent the mapping relationship between multi-dimensional scheduling feature information and vehicle scheduling decisions; S4: Obtain real-time traffic information for the target area, and perform dynamic path optimization on the preliminary scheduling plan based on the real-time traffic information to generate the optimal driving path for each available vehicle. S5: Generates scheduling instructions based on the optimal driving route and sends the scheduling instructions to the corresponding available vehicles, controlling the available vehicles to perform transportation tasks.

2. The intelligent vehicle scheduling method integrating AI multi-dimensional perception and dynamic optimization according to claim 1, characterized in that, The real-time operational data includes the vehicle's current location information, vehicle load status information, and vehicle operational status information. The step in S2, which extracts features from real-time running data, starting point location information, and ending point location information to generate multi-dimensional scheduling feature information, includes: Spatial distance analysis is performed between the vehicle's current location information and its starting point location information to obtain the first feature data; A matching degree analysis is performed between the vehicle load status information and the cargo information of the transportation task to obtain the second feature data; The steps for performing a matching degree analysis between vehicle load status information and cargo information of transportation tasks to obtain second feature data include: comparing the vehicle's rated load with the cargo weight of the transportation task; when the vehicle's remaining load is greater than the cargo weight, calculating the normalized score of the difference between the two as the second feature data. The available time of a vehicle is determined based on its operating status information, and the first feature data, the second feature data, and the available time are integrated to generate multi-dimensional scheduling feature information.

3. The intelligent vehicle scheduling method integrating AI multi-dimensional perception and dynamic optimization according to claim 1, characterized in that, The step in S4, which performs dynamic path optimization on the preliminary scheduling plan based on real-time traffic information to generate the optimal travel path for each available vehicle, includes: The task sequence and corresponding waypoint information for each available vehicle are determined based on the preliminary scheduling plan. Obtain real-time traffic flow and traffic incident information within the target area as real-time traffic information; The information on waypoints is integrated with real-time traffic information to generate a dynamic road network model. The steps for generating the dynamic road network model include: dynamically adjusting the road weight values ​​of each road segment in the dynamic road network model based on real-time traffic flow information, wherein the road weight value of congested road segments is increased and the road weight value of unobstructed road segments is decreased. The driving paths in the task sequence are iteratively optimized based on the dynamic road network model to generate the optimal driving path for each available vehicle to avoid congested road sections.

4. The intelligent vehicle scheduling method integrating AI multi-dimensional perception and dynamic optimization according to claim 1, characterized in that, Before the step of obtaining the starting location information, ending location information, and real-time operation data of multiple available vehicles for multiple transportation tasks within the target area in step S1, the method further includes: Collect historical transportation task data and corresponding historical scheduling scheme data to construct a training sample set; Using historical transportation task data from the training sample set as input and historical scheduling scheme data as supervision labels, the initial neural network model is trained to obtain the scheduling decision model.

5. The intelligent vehicle scheduling method integrating AI multi-dimensional perception and dynamic optimization according to claim 1, characterized in that, After the step in S5 of generating a dispatch instruction based on the optimal driving route, sending the dispatch instruction to the corresponding available vehicle, and controlling the available vehicle to execute the transportation task, the method further includes: Real-time monitoring of the actual driving trajectory of available vehicles during the transportation task; When the actual driving trajectory is detected to deviate from the optimal driving path and the deviation distance exceeds a preset threshold, a path replanning instruction is triggered. Based on the current location of available vehicles and real-time traffic information, an updated optimal driving route is regenerated.

6. An intelligent vehicle dispatching system integrating AI multi-dimensional perception and dynamic optimization, used to implement the method of any one of claims 1-5, characterized in that, include: The data acquisition module is used to acquire the starting point location information, ending point location information, and real-time operation data of multiple available vehicles for multiple transportation tasks within the target area; The feature extraction module is used to extract features from real-time running data, as well as start-up and end-up location information, to generate multi-dimensional scheduling feature information. The scheme generation module is used to input multi-dimensional scheduling feature information into a pre-trained scheduling decision model and output preliminary scheduling schemes for multiple transportation tasks; wherein, the scheduling decision model is constructed based on a machine learning model and is used to represent the mapping relationship between multi-dimensional scheduling feature information and vehicle scheduling decisions; The route optimization module is used to obtain real-time traffic information of the target area and perform dynamic route optimization on the preliminary scheduling plan based on the real-time traffic information to generate the optimal driving route for each available vehicle. The instruction execution module is used to generate scheduling instructions based on the optimal driving route and send the scheduling instructions to the corresponding available vehicles to control the available vehicles to perform transportation tasks.

7. The intelligent vehicle dispatching system integrating AI multi-dimensional perception and dynamic optimization according to claim 6, characterized in that, The real-time operational data includes the vehicle's current location information, vehicle load status information, and vehicle operational status information. The feature extraction module includes: The first analysis unit is used to perform spatial distance analysis on the vehicle's current location information and the starting point location information to obtain the first feature data; The second analysis unit is used to perform a matching degree analysis on the vehicle load status information and the cargo information of the transportation task to obtain the second feature data. The second feature data includes: comparing the vehicle's rated load with the cargo weight of the transportation task. When the vehicle's remaining load is greater than the cargo weight, the normalized score of the difference between the two is calculated as the second feature data. The data integration unit is used to determine the available time of a vehicle based on the vehicle's operating status information, and to integrate the first feature data, the second feature data, and the available time to generate multi-dimensional scheduling feature information.

8. The intelligent vehicle dispatching system integrating AI multi-dimensional perception and dynamic optimization according to claim 6, characterized in that, The path optimization module includes: The task parsing unit is used to determine the task sequence and corresponding waypoint information for each available vehicle based on the preliminary scheduling plan. The traffic data acquisition unit is used to acquire real-time traffic flow information and traffic event information within the target area, as real-time traffic information. The model fusion unit is used to fuse waypoint information with real-time traffic information to generate a dynamic road network model. The generation of the dynamic road network model includes: dynamically adjusting the road right-of-way value of each road segment in the dynamic road network model based on real-time traffic flow information, wherein the road right-of-way value of congested road segments is increased and the road right-of-way value of unobstructed road segments is decreased. The path iteration unit is used to iteratively optimize the driving paths in the task sequence based on the dynamic road network model, and generate the optimal driving path for each available vehicle to avoid congested road sections.

9. The intelligent vehicle dispatching system integrating AI multi-dimensional perception and dynamic optimization according to claim 6, characterized in that, The system also includes a model training module, which is used to collect historical transportation task data and corresponding historical scheduling scheme data to construct a training sample set before the data acquisition module acquires the starting location information, ending location information and real-time operation data of multiple transportation tasks in the target area and multiple available vehicles. Using historical transportation task data from the training sample set as input and historical scheduling scheme data as supervision labels, the initial neural network model is trained to obtain the scheduling decision model.

10. The intelligent vehicle dispatching system integrating AI multi-dimensional perception and dynamic optimization according to claim 6, characterized in that, The system also includes a dynamic adjustment module, which is used to monitor the actual driving trajectory of the available vehicles in real time after the instruction execution module generates a scheduling instruction based on the optimal driving route and sends the scheduling instruction to the corresponding available vehicles and controls the available vehicles to perform the transportation task. When the actual driving trajectory is detected to deviate from the optimal driving path and the deviation distance exceeds a preset threshold, a path replanning instruction is triggered; based on the current location of available vehicles and real-time traffic information, an updated optimal driving path is regenerated.

Images