Scheduling method, device and equipment of automatic driving vehicle and storage medium

Abstract

The invention provides an automatic driving vehicle scheduling method and device, equipment and a storage medium, and the method comprises the steps: carrying out the multi-dimensional collection and preprocessing of an automatic driving vehicle, and generating a comprehensive data set; and based on the comprehensive data set, carrying out dynamic allocation and fleet collaborative planning on loading and unloading tasks to generate a task allocation scheme. And through a hierarchical search algorithm and a multi-objective optimization algorithm, the path of the vehicle is planned and optimized in combination with the task allocation scheme and the comprehensive data set, and an optimized path scheme is generated. And according to the task allocation scheme, the optimized path scheme, the integrated data set and the vehicle real-time state data, utilizing a predictive analysis model to evaluate and deal with potential risks in the transportation process, and generating and executing a real-time scheduling instruction. According to the invention, the scheduling method of multi-dimensional data acquisition and processing, dynamic task allocation, path optimization and risk prediction is combined, the scheduling efficiency and safety of the autonomous vehicle are comprehensively improved, and efficient logistics management is realized.

Description

Technical Field

[0001] The present invention relates to the field of vehicle dispatching, and in particular to a dispatching method, device, equipment and storage medium for an autonomous driving vehicle. Background Art

[0002] With the widespread application of autonomous driving technology in the logistics field, the scheduling method of autonomous driving vehicles has become the key to achieving efficient transportation and intelligent logistics management. Traditional logistics scheduling mainly relies on manual experience and rule-driven, and completes transportation needs through fixed path planning and static task allocation. However, this method is less efficient when facing complex logistics scenarios, especially in multi-fleet collaboration and dynamic task allocation, and it is difficult to give full play to the flexibility and intelligence advantages of autonomous driving vehicles.

[0003] Existing scheduling methods usually optimize for a single goal, such as the shortest path or optimal time, and lack comprehensive consideration of cargo characteristics, environmental conditions, and the real-time status of the fleet. Especially in complex dynamic environments, such as traffic congestion, weather changes, or changes in cargo loading and unloading requirements, existing methods are difficult to quickly adjust scheduling plans, resulting in reduced transportation efficiency or increased costs. In addition, traditional methods also have shortcomings in risk assessment and response, and are unable to make accurate predictions and rapid responses through real-time data, which increases uncertainty in the transportation process. Summary of the invention

[0004] The main purpose of the present invention is to solve the technical problems in existing logistics scheduling methods, such as the inability to flexibly cope with complex dynamic environments, the lack of multi-objective optimization and fleet collaborative planning capabilities, and the insufficient response in risk assessment and real-time adjustment; A first aspect of the present invention provides a method for dispatching an autonomous driving vehicle, the method comprising: The truck operation data, road information data, cargo characteristics data and environmental condition data of the autonomous driving vehicle are collected and preprocessed in multiple dimensions to obtain a preprocessed comprehensive data set; Using a preset cargo classification model and a rolling horizon optimization strategy, based on the comprehensive data set, dynamically allocate preset loading and unloading tasks and coordinate fleet planning to obtain a task allocation plan; By using a preset hierarchical search algorithm and a multi-objective optimization algorithm, a truck path of the autonomous driving vehicle is planned and optimized according to the task allocation scheme and the comprehensive data set to obtain an optimized path plan; Based on the task allocation plan, optimized path plan, comprehensive data set and real-time status data of the truck of the autonomous driving vehicle, a predictive analysis model is used to evaluate and respond to potential risks in the transportation process, and real-time scheduling instructions are generated and executed.

[0005] Optionally, in a first implementation of the first aspect of the present invention, the preset cargo classification model and the rolling horizon optimization strategy are used to dynamically allocate preset loading and unloading tasks and coordinate fleet planning based on the comprehensive data set to obtain a task allocation plan, which includes: Performing cluster analysis on the cargo characteristic data in the comprehensive data set to obtain a cargo category matrix, and using an improved Hungarian algorithm to perform preliminary cargo-vehicle matching according to the cargo category matrix and the load capacity of the autonomous driving vehicle to obtain an initial allocation plan; Applying a task merging algorithm based on time and space constraints to the initial allocation plan, integrating the loading and unloading tasks that meet the conditions to obtain an optimized task combination; Using a dynamic programming algorithm, according to the optimized task combination and the truck operation data, the optimal loading and unloading sequence is calculated to obtain a preliminary scheduling sequence; The preliminary scheduling sequence is dynamically updated within a preset time window by using a rolling time domain optimization strategy, and new tasks are inserted into the preliminary scheduling sequence in real time to obtain a task allocation plan.

[0006] Optionally, in a second implementation of the first aspect of the present invention, applying a task merging algorithm based on spatiotemporal constraints to the initial allocation plan to integrate qualified loading and unloading tasks to obtain an optimized task combination includes: Performing time-space coordinate mapping on the loading and unloading tasks in the initial allocation plan to obtain a task time-space distribution matrix; Using a density clustering algorithm to perform cluster analysis on the task spatiotemporal distribution matrix, a potential mergeable task set is obtained; According to a preset time window and space distance threshold, the potentially merging task set is screened to obtain task pairs that meet the merging conditions; The maximum weight matching algorithm in graph theory is used to optimally combine the task pairs that meet the merging conditions to obtain a preliminary merging solution; A heuristic search algorithm is applied to the preliminary merging scheme, and local optimization adjustment is performed considering vehicle capacity and path constraints to obtain an optimized task combination.

[0007] Optionally, in a third implementation of the first aspect of the present invention, the truck path of the autonomous driving vehicle is planned and optimized according to the task allocation scheme and the comprehensive data set by using a preset hierarchical search algorithm and a multi-objective optimization algorithm to obtain an optimized path plan, including: According to the road information data in the comprehensive data set, a multi-level road network model is constructed, and a coarse-grained path search is performed in the multi-level road network model using an improved Dijkstra algorithm to obtain a candidate path set, wherein the multi-level road network model is a hierarchical graph structure including trunk roads and secondary roads; Applying the A* algorithm that takes into account the turning radius and height restrictions of the truck to the candidate path set, performing fine-grained path search, and obtaining a feasible path set that meets the constraint conditions; According to the time window requirement in the task allocation scheme, the feasible path set is expanded in time dimension, a space-time network diagram is constructed, and a multi-objective optimization method based on a genetic algorithm is applied to the space-time network diagram to obtain a Pareto optimal path set; According to the environmental condition data and the truck operation data in the comprehensive data set, the most suitable path is selected from the Pareto optimal path set as the optimized path solution.

[0008] Optionally, in a fourth implementation of the first aspect of the present invention, according to the time window requirement in the task allocation scheme, the feasible path set is expanded in time dimension, a space-time network graph is constructed, and a multi-objective optimization method based on a genetic algorithm is applied to the space-time network graph to obtain a Pareto optimal path set, including: Discrete time sampling is performed on each path in the feasible path set according to the time window requirement in the task allocation scheme to obtain a time-expanded path node set; Using a dynamic time warping algorithm, aligning and interpolating the time-expanded path node set to obtain a regularized space-time path; Constructing a directed acyclic graph based on the space-time path, and applying a topological sorting algorithm to generate a space-time network graph that satisfies the timing constraints; Initializing a genetic algorithm population according to a preset multi-objective fitness function and the spatiotemporal network graph, and performing crossover, mutation and selection operations on the genetic algorithm population, and iteratively optimizing to obtain a non-dominated solution set; The non-dominated solution set is sorted and screened using the NSGA-II algorithm to obtain a Pareto optimal path set.

[0009] Optionally, in a fifth implementation of the first aspect of the present invention, the step of evaluating and responding to potential risks in the transportation process using a predictive analysis model based on the task allocation plan, the optimized path plan, the comprehensive data set, and the real-time status data of the truck of the autonomous driving vehicle, and generating and executing real-time scheduling instructions includes: Extracting and fusing features of the comprehensive data set and the real-time status data of the truck to obtain a multi-dimensional feature vector, and inputting the multi-dimensional feature vector into a pre-trained time series prediction model to obtain a prediction result set; Using a multi-criteria decision analysis method to quantitatively evaluate potential risks based on the prediction result set, generate a risk score, and use a heuristic algorithm to dynamically adjust the task allocation plan and the optimized path plan according to the risk score to obtain an adjusted scheduling strategy; The adjusted scheduling strategy is converted into specific vehicle instructions and task instructions to obtain a real-time scheduling instruction set, the real-time scheduling instruction set is distributed to the corresponding autonomous driving vehicle, and the execution of the real-time scheduling instruction set is monitored to achieve real-time scheduling control.

[0010] Optionally, in a sixth implementation of the first aspect of the present invention, the multi-criteria decision analysis method is used to quantitatively evaluate potential risks based on the prediction result set to generate a risk score, and a heuristic algorithm is used to dynamically adjust the task allocation plan and the optimized path plan according to the risk score, and the adjusted scheduling strategy includes: Constructing a decision matrix based on the prediction result set, and calculating the score of each potential risk using a multi-criteria decision analysis method and the decision matrix to obtain a risk score set; Determine the tasks and paths that need to be adjusted according to the risk score set, and generate an adjustment target list; Using a heuristic algorithm to reallocate and adjust the paths of the tasks in the adjustment target list to generate multiple candidate scheduling solutions; The risk level and scheduling efficiency of the candidate scheduling solutions are evaluated, and the optimal solution is selected as the adjusted scheduling strategy.

[0011] A second aspect of the present invention provides a dispatching device for an autonomous driving vehicle, the dispatching device for the autonomous driving vehicle comprising: A data processing module is used to collect and preprocess the truck operation data, road information data, cargo characteristics data and environmental condition data of the autonomous driving vehicle in multiple dimensions to obtain a preprocessed comprehensive data set; A task allocation module, which is used to dynamically allocate preset loading and unloading tasks and coordinate fleet planning based on the comprehensive data set by using a preset cargo classification model and a rolling time domain optimization strategy to obtain a task allocation plan; A path optimization module, for planning and optimizing the truck path of the autonomous driving vehicle according to the task allocation scheme and the comprehensive data set through a preset hierarchical search algorithm and a multi-objective optimization algorithm to obtain an optimized path plan; The scheduling module is used to evaluate and respond to potential risks in the transportation process using a predictive analysis model based on the task allocation plan, optimized path plan, comprehensive data set and real-time status data of the truck of the autonomous driving vehicle, and generate and execute real-time scheduling instructions.

[0012] A third aspect of the present invention provides a scheduling device for an autonomous driving vehicle, comprising: a memory and at least one processor, wherein instructions are stored in the memory, and the memory and the at least one processor are interconnected via lines; the at least one processor calls the instructions in the memory so that the scheduling device of the autonomous driving vehicle executes the steps of the above-mentioned scheduling method for the autonomous driving vehicle.

[0013] A fourth aspect of the present invention provides a computer-readable storage medium, which stores instructions that, when executed on a computer, enable the computer to execute the steps of the above-mentioned method for dispatching an autonomous driving vehicle.

[0014] The above-mentioned scheduling method, device, equipment and storage medium for autonomous driving vehicles generate a comprehensive data set by multi-dimensional collection and preprocessing of autonomous driving vehicles. Based on the comprehensive data set, the loading and unloading tasks are dynamically allocated and the fleet is collaboratively planned to generate a task allocation plan. Through a hierarchical search algorithm and a multi-objective optimization algorithm, the vehicle's path is planned and optimized in combination with the task allocation plan and the comprehensive data set, and an optimized path plan is generated. According to the task allocation plan, the optimized path plan, the comprehensive data set and the real-time status data of the vehicle, a predictive analysis model is used to evaluate and respond to potential risks in the transportation process, and real-time scheduling instructions are generated and executed. The present invention combines multi-dimensional data collection and processing, dynamic task allocation, path optimization and risk prediction scheduling methods to comprehensively improve the scheduling efficiency and safety of autonomous driving vehicles and achieve efficient logistics management.

[0015] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or understood by practicing the present invention. The purpose and other advantages of the present invention are realized and obtained by the structures particularly pointed out in the description, claims and drawings.

[0016] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 Schematic diagram of a first embodiment of a method for dispatching an autonomous driving vehicle in an embodiment of the present invention; Figure 2 A schematic diagram of an embodiment of a dispatching device for an autonomous driving vehicle in an embodiment of the present invention; Figure 3 Schematic diagram of an embodiment of a dispatching device for an autonomous driving vehicle in an embodiment of the present invention. DETAILED DESCRIPTION

[0018] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the described embodiments are 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.

[0019] The terms "including" and "having" and any variations thereof mentioned in the embodiments of the present invention are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device end including a series of steps or units is not limited to the listed steps or units, but may optionally include other steps or units that are not listed, or may optionally include other steps or units that are inherent to these processes, methods, products or device ends.

[0020] To facilitate understanding of this embodiment, a method for dispatching an autonomous driving vehicle disclosed in an embodiment of the present invention is first described in detail. Figure 1 As shown, the method comprises the following steps: 101. Collect and preprocess the truck operation data, road information data, cargo characteristics data and environmental condition data of the autonomous driving vehicle in multiple dimensions to obtain a preprocessed comprehensive data set; In one embodiment of the present invention, a multi-source data acquisition system is firstly established. The system collects the real-time position, speed, load, engine status and tire pressure of the truck through the on-board sensors. At the same time, through cooperation with the transportation department, road information data such as real-time road conditions, traffic flow and accident information are obtained. The cargo characteristic data is collected through the logistics management system, including cargo type, weight, volume and special requirements. The environmental condition data is obtained through the on-board meteorological sensors and the public meteorological data interface, including information such as temperature, humidity, wind speed and precipitation. After the data collection is completed, the system performs preliminary data cleaning to remove obvious outliers and erroneous data. Subsequently, the data from different sources are synchronized in time and aligned in space to ensure the consistency of the data. Next, interpolation and moving average methods are used to handle missing values ​​and outliers, such as filling missing traffic flow data or smoothing abnormal speed data. Then, the data is standardized and normalized to convert data of different scales into a unified range. Finally, the system analyzes the historical data to extract valuable features, such as the periodic pattern of traffic flow and the time distribution characteristics of customer orders. After these steps, a comprehensive dataset containing truck operating status, road traffic conditions, cargo information and environmental conditions is finally obtained.

[0021] 102. Using the preset cargo classification model and rolling time domain optimization strategy based on the comprehensive data set, dynamically allocate the preset loading and unloading tasks and coordinate fleet planning to obtain the task allocation plan; In one embodiment of the present invention, the preset cargo classification model and rolling time domain optimization strategy are used to dynamically allocate preset loading and unloading tasks and coordinate fleet planning based on the comprehensive data set to obtain a task allocation plan, including: clustering analysis of the cargo characteristic data in the comprehensive data set to obtain a cargo category matrix, and using an improved Hungarian algorithm to perform preliminary cargo-vehicle matching based on the cargo category matrix and the load capacity of the autonomous driving vehicle to obtain an initial allocation plan; applying a task merging algorithm based on spatiotemporal constraints to the initial allocation plan to integrate qualified loading and unloading tasks to obtain an optimized task combination; using a dynamic programming algorithm to calculate the optimal loading and unloading sequence based on the optimized task combination and the truck operation data to obtain a preliminary scheduling sequence; using a rolling time domain optimization strategy to dynamically update the preliminary scheduling sequence within a preset time window, and insert new tasks into the preliminary scheduling sequence in real time to obtain a task allocation plan.

[0022] Specifically, first, the system extracts cargo characteristic data from the comprehensive data set, including information on multiple dimensions such as weight, volume, timeliness, special requirements (such as temperature control, shockproof), etc. Then, the K-means clustering algorithm is used to perform cluster analysis on these characteristic data. The K-means algorithm classifies similar cargo into the same category through iterative calculation to form a cargo category matrix. This matrix contains the center point of each cargo category (i.e., the average characteristic value) and information about which category each cargo belongs to. Next, the system uses the improved Hungarian algorithm for preliminary cargo-vehicle matching. The traditional Hungarian algorithm is used to solve the maximum weight matching problem of bipartite graphs, while the improved algorithm takes into account the vehicle's load capacity and the diversity of cargo. The algorithm first constructs a cost matrix, in which each element represents the cost of assigning a certain type of cargo to a certain vehicle (considering factors such as load matching degree and vehicle characteristics). Then, through iterative optimization, the allocation scheme with the lowest total cost is found. This process requires multiple adjustments to ensure that the allocation task of each vehicle does not exceed its load capacity, while meeting the transportation needs of all cargo as much as possible. Ultimately, the system obtains an initial allocation plan that contains information about what cargo each autonomous vehicle should load, providing a basis for subsequent optimization.

[0023] Specifically, after obtaining the initial allocation plan, the system applies a task merging algorithm based on time and space constraints to improve transportation efficiency and reduce costs. First, the algorithm maps the time and space coordinates of each loading and unloading task in the initial allocation plan to create a task time and space distribution matrix. This matrix contains the geographical location and expected execution time of each task. Then, DBSCAN (density-based spatial clustering algorithm) is used to perform cluster analysis on these tasks to identify task groups that are similar in time and space. The algorithm sets time and distance thresholds, and only considers merging two tasks when the execution time difference between them is within the allowable range and the geographical location is close enough. Then, for each potentially mergeable task pair, the total distance, time and cost after merging are calculated and compared with the case without merging. If the merger can bring significant benefits, the pair of tasks is marked as mergeable. After that, the maximum weight matching algorithm in graph theory is used to find the optimal merger combination among all mergeable task pairs. This process needs to take into account vehicle capacity constraints to ensure that the merged tasks do not exceed the vehicle's carrying capacity. Finally, the algorithm performs local optimization on the merger plan and may adjust some merger decisions to balance overall efficiency. Through this process, the system obtains an optimized task combination, reducing the vehicle's empty rate and total driving distance while ensuring that the time window is met.

[0024] Specifically, after obtaining the optimized task combination, the system needs to determine the optimal loading and unloading sequence for each autonomous truck to maximize efficiency and meet various constraints. This step is solved using a dynamic programming algorithm. First, the system treats the task of each truck as a subproblem, with the goal of finding the shortest path to complete all assigned tasks. The dynamic programming algorithm starts from the last task and works forward step by step. For each task point, the algorithm calculates the optimal path from that point to the end point and stores this information. Multiple factors are considered in the calculation process, including the distance between tasks, time window constraints, the current load of the truck, road conditions, etc. This information comes from previous comprehensive data sets and truck operation data. The algorithm uses a state transition equation that takes into account the cost (time and distance) of moving from one task point to the next, as well as the time required to complete the task. By repeatedly applying this equation, the algorithm gradually constructs the optimal path from the starting point to the end point. When considering the time window constraint, if a path causes the task to be unable to be completed within the specified time, the path will be excluded. In addition, the algorithm also needs to handle the logical constraints of the loading and unloading sequence to ensure that the loading and unloading sequence of the goods is reasonable. Ultimately, for each truck, the algorithm outputs a preliminary dispatch sequence that contains all the tasks that the truck needs to perform, arranged in the optimal order while satisfying the time window and other constraints.

[0025] Specifically, a fixed-length time window is first set, such as the next 4 hours. Within this time window, the system continuously monitors various real-time data, including traffic conditions, weather changes, and new orders. At regular intervals (for example, 15 minutes), the system triggers an optimization process. In each optimization, the system first evaluates the progress of the currently executing scheduling sequence and identifies completed tasks and ongoing tasks. Then, the system merges the newly added tasks with the tasks that have not yet started in the time window to form a new set of pending tasks. For this new set of tasks, the system reruns the previous optimization steps, including task classification, vehicle matching, and path planning. When inserting a new task, the system evaluates multiple insertion locations and selects the location with the least impact on the overall scheduling. This process uses heuristic algorithms, such as simulated annealing or genetic algorithms, to quickly find a better insertion plan. The optimization process also needs to consider tasks that are already in execution to avoid causing excessive interference to these tasks. If the new optimization result is significantly better than the currently executed plan, the system will generate adjustment instructions to guide the vehicle to execute according to the new plan. Through this continuous dynamic optimization, the system can respond to various changes in a timely manner, including sudden orders, traffic congestion, etc., thereby maintaining the efficiency and flexibility of the scheduling plan. The resulting task allocation solution is a dynamically updated plan that balances efficiency and adaptability, providing continuously optimized scheduling guidance for autonomous truck fleets.

[0026] Furthermore, the initial allocation plan is applied with a task merging algorithm based on time-space constraints to integrate the qualified loading and unloading tasks to obtain an optimized task combination, including: performing time-space coordinate mapping on the loading and unloading tasks in the initial allocation plan to obtain a task time-space distribution matrix; performing cluster analysis on the task time-space distribution matrix using a density clustering algorithm to obtain a set of potentially mergeable tasks; screening the set of potentially mergeable tasks according to a preset time window and spatial distance threshold to obtain task pairs that meet the merging conditions; using the maximum weight matching algorithm in graph theory to optimally combine the task pairs that meet the merging conditions to obtain a preliminary merging plan; applying a heuristic search algorithm to the preliminary merging plan, considering vehicle capacity and path constraints, performing local optimization adjustments to obtain an optimized task combination.

[0027] Specifically, it is necessary to first extract the key information of each loading and unloading task from the initial allocation plan. This information includes the geographical coordinates (longitude and latitude) of the task, the expected execution time and the time window. In order to create the spatiotemporal distribution matrix, the system combines the time and space dimensions. In the spatial dimension, a two-dimensional coordinate system is used to represent the geographical location of the task. The time dimension is regarded as the third dimension, forming a three-dimensional space. Each task is represented as a point or a small time period in this three-dimensional space (considering that the task execution may take a certain amount of time). The system also needs to consider the constraints of the time window and represent it as a range in the time dimension. In order to facilitate subsequent processing, the system discretizes the continuous spatiotemporal coordinates, divides the time into fixed time periods (for example, 15 minutes as a unit), and the spatial coordinates are gridded according to the actual geographical range. In this way, each task is mapped to one or more discrete spatiotemporal units. The resulting task spatiotemporal distribution matrix is ​​a multidimensional array, in which each element represents a spatiotemporal unit, and the value of the element represents the number of tasks or task characteristics in the unit. This matrix not only contains the spatial distribution information of the task, but also reflects the distribution and concentration of the task in time, providing a basic data structure for subsequent clustering analysis.

[0028] Specifically, the density clustering algorithm is used to perform cluster analysis on the tasks, and the DBSCAN (Density-based Spatial Clustering with Applied Noise) algorithm is selected because it can discover clusters of arbitrary shapes and has good robustness to noise points. The DBSCAN algorithm requires two key parameters: ε (Epsilon) defines the radius of the neighborhood, and MinPts defines the minimum number of neighbors required to become a core point. The algorithm first regards each non-empty element in the spatiotemporal distribution matrix as a point, and then traverses all points. For each point, the algorithm calculates the number of points in its ε-neighborhood. If the number is greater than or equal to MinPts, the point is marked as a core point and a new cluster is formed. Then, the algorithm recursively adds all density-reachable points to this cluster. Points that do not belong to any cluster are considered noise points. In a spatiotemporal environment, the ε parameter needs to consider both time and space dimensions, and the weighted Euclidean distance can be used to calculate the distance between points. The MinPts parameter is set according to the desired clustering density. After the clustering process is completed, the system obtains a series of spatiotemporal clusters, each of which represents a group of tasks that are close in time and space. These clusters form a set of potentially mergeable tasks. Tasks within each cluster are potential objects that can be merged because they are close enough in time and space. The system also calculates some statistical features for each cluster, such as the spatiotemporal range of the cluster, the number of tasks, the distribution of task types, etc. This information will be used for subsequent merging condition screening.

[0029] Specifically, after obtaining the set of potentially mergeable tasks, the system needs to further filter out the task pairs that truly meet the merging conditions. First, it needs to determine the appropriate time window and spatial distance threshold. The time window threshold defines the maximum allowable time difference between two tasks, while the spatial distance threshold defines the maximum allowable distance between tasks in geographical locations. The setting of these thresholds needs to take into account actual operating conditions, such as vehicle speed, traffic conditions, loading and unloading time, and other factors. The screening process starts from within each cluster. The system will traverse each pair of tasks in the cluster and calculate the time difference and spatial distance between them. The calculation of the time difference needs to consider the execution time window of the task to ensure that the merged tasks can still be completed within their respective time windows. The calculation of the spatial distance needs to consider the actual road network distance, not just the straight-line distance. If the time difference between a pair of tasks is less than the preset time window threshold, and the spatial distance is less than the preset spatial distance threshold, the pair of tasks is marked as a task pair that meets the merging conditions. During the screening process, the system also needs to consider other attributes of the task, such as the compatibility of cargo types, the load limit of the vehicle, and so on. For each task pair that meets the conditions, the system will calculate the potential benefits after merging, such as the total driving distance saved, the improved vehicle utilization rate, and so on. This information will be used in the subsequent optimal combination process. Finally, the system obtains a list of task pairs that meet the merging conditions. Each task pair contains two tasks that can be merged and their related merged benefit information.

[0030] Specifically, after obtaining a list of task pairs that meet the merging conditions, the system needs to select the optimal combination solution from them. This problem can be transformed into the maximum weight matching problem in graph theory. First, the system constructs an undirected graph in which each node represents a task, an edge indicates that two tasks can be merged, and the weight of the edge represents the benefit after the merger (such as the saved time or distance). Then, the system applies the Edmonds' blossom algorithm to solve this maximum weight matching problem. The core idea of ​​this algorithm is to increase the weight sum of the matching by continuously searching for augmenting paths. The algorithm starts with an empty matching and then iteratively searches for augmenting paths that can increase the total weight. In the process of searching for augmenting paths, the algorithm needs to deal with the "flower" structure, which is the feature and difficulty of the algorithm. When encountering an odd ring, the algorithm shrinks the entire ring into a super vertex and then continues to search for augmenting paths. After finding an augmenting path, the algorithm updates the current matching. This process is repeated until no more augmenting paths can be found. The time complexity of the Edmonds' blossom algorithm is O(n^3), where n is the number of nodes in the graph. In practical applications, some heuristic methods can be used to speed up the algorithm, such as using a greedy strategy to quickly find an initial match, or using data structures such as Fibonacci heaps to optimize certain operations of the algorithm. After the algorithm is executed, the maximum weight match obtained is the preliminary task merging plan. This plan ensures that the total benefit after merging is maximized under given constraints. However, this preliminary plan may need further optimization because it does not take into account some global constraints, such as the overall path planning of the vehicle.

[0031] Specifically, although the preliminary merging scheme optimizes the task combination locally, it may not fully consider the global constraints, such as vehicle capacity constraints and overall path planning. Therefore, further optimization is needed. This step uses a heuristic search algorithm, and specifically the simulated annealing algorithm can be used for local optimization adjustments. First, the system uses the preliminary merging scheme as the initial solution. Then, an objective function is defined that comprehensively considers factors such as total driving distance, vehicle utilization, and task completion time. The simulated annealing algorithm searches for a better solution through iteration. In each iteration, the algorithm randomly selects a neighborhood operation, such as exchanging the assignment of two tasks, moving a task from one vehicle to another, etc. This new solution will be evaluated, and if it is better than the current solution, it will be accepted; if it is slightly worse than the current solution, there is also a certain probability of being accepted, and this probability decreases with the increase in the number of iterations (simulated annealing process). This strategy allows the algorithm to jump out of the local optimum during the search process. When evaluating the new solution, the algorithm needs to check whether the vehicle capacity constraints and path feasibility are met. If these constraints are violated, the new solution will be rejected or penalized. The algorithm also considers the smoothness of the path to avoid generating overly complex paths. As the "temperature" decreases, the algorithm tends to accept better solutions more and more, and eventually converges to a high-quality solution.

[0032] 103. Through the preset hierarchical search algorithm and multi-objective optimization algorithm, the truck path of the autonomous driving vehicle is planned and optimized according to the task allocation plan and the comprehensive data set to obtain the optimized path plan; In one embodiment of the present invention, the truck path of the autonomous driving vehicle is planned and optimized according to the task allocation scheme and the comprehensive data set by using a preset hierarchical search algorithm and a multi-objective optimization algorithm to obtain an optimized path plan, including: constructing a multi-level road network model according to the road information data in the comprehensive data set, and using an improved Dijkstra algorithm to perform a coarse-grained path search in the multi-level road network model to obtain a candidate path set, wherein the multi-level road network model is a hierarchical graph structure including trunk roads and secondary roads; applying an A* algorithm that takes into account the turning radius and height restrictions of the truck to the candidate path set to perform a fine-grained path search to obtain a feasible path set that meets the constraint conditions; according to the time window requirements in the task allocation scheme, the feasible path set is expanded in time dimension to construct a spatiotemporal network diagram, and a multi-objective optimization method based on a genetic algorithm is applied to the spatiotemporal network diagram to obtain a Pareto optimal path set; according to the environmental condition data in the comprehensive data set and the truck operation data, the most suitable path is selected from the Pareto optimal path set as the optimized path plan.

[0033] Specifically, the multi-level road network model is constructed to more effectively process large-scale road network data and improve the efficiency of path search. First, the system extracts road information data from the comprehensive data set, including road type, length, capacity and other attributes. Then, the road network is divided into two levels: trunk roads and secondary roads according to the road level. The trunk road layer includes important traffic arteries such as expressways and national roads, and the secondary road layer includes lower-level roads such as provincial roads and county roads. During the construction process, the system assigns a unique identifier to each road and records the node information connected to it. For cross-level connection points, such as expressway exits, the system creates special conversion nodes for switching between different levels. This hierarchical structure allows the algorithm to quickly plan a rough path at a high level and then refine it at a low level. Next, the system applies the improved Dijkstra algorithm for coarse-grained path search. The improvements are mainly reflected in two aspects: first, the algorithm prioritizes searching the high-level road network and only descends to the low level when necessary; second, the heuristic function is introduced to make the algorithm tilt the search in the target direction. In the specific implementation, the algorithm maintains a priority queue and takes out the node with the smallest estimated total cost from the queue for expansion each time. The estimated total cost includes the actual cost from the starting point to the current node and the estimated cost from the current node to the end point. This method significantly reduces the search space and improves the efficiency of the algorithm. During the search process, the algorithm records multiple possible paths to form a candidate path set. This set contains multiple approximate routes from the starting point to the end point, providing a basis for subsequent fine-grained searches.

[0034] Specifically, after obtaining the candidate path set, a more refined path search is required to ensure that the path meets the special needs of trucks, so the A algorithm that considers the turning radius and height restrictions of trucks is used. First, the system extracts truck-related parameters such as vehicle length, width, height, and minimum turning radius from the comprehensive data set. Then, for each candidate path, the system performs a fine-grained search on the low-level road network. The core of the A algorithm is to design appropriate heuristic functions and cost functions. The heuristic function estimates the distance from the current node to the target node, usually using straight-line distance or Manhattan distance. The cost function needs to take into account multiple factors: road length, estimated travel time, turning difficulty, and height restrictions. For each road segment, the system checks whether its width and turning radius meet the truck requirements. If not, the segment will be assigned an extremely high cost or directly excluded. For bridges or tunnels, the system checks whether its height limit meets the truck requirements. During the search process, the algorithm also needs to consider the situation of continuous turns to ensure that the generated path has enough operating space at the turn. To improve efficiency, the algorithm uses a hierarchical search strategy: first quickly plan a rough route on the main road layer, and then switch to the secondary road layer for detailed search in key areas (such as near the start and end points or complex sections). The search results are a set of feasible paths that meet the truck's turning radius and height restrictions. These paths not only take into account distance and time factors, but also ensure that the truck can pass safely. The system calculates detailed turning instructions and height warning information for each feasible path, providing a basis for subsequent path selection and driving guidance.

[0035] Specifically, the set of feasible paths is extended to the time dimension to meet the time window requirements in the task allocation scheme while considering the time-varying characteristics of traffic flow. First, the system discretizes each feasible path on the time axis according to the time window of the task, usually with 5 minutes or 10 minutes as a time unit. For each node on the path, the system creates multiple copies of time points to form a spatiotemporal node. These spatiotemporal nodes are connected by spatiotemporal edges, and the weight of the edge represents the cost of moving from one node to another in a specific time period (such as travel time, energy consumption, etc.). The system also needs to consider the situation of staying in the same place, which is represented by vertical edges. In this way, the original spatial road network is expanded into a spatiotemporal network graph. Next, the system applies a multi-objective optimization method based on genetic algorithm on this spatiotemporal network graph. The optimization objectives include total travel time, energy consumption, punctuality, etc. The chromosome encoding scheme of the genetic algorithm adopts path representation, and each gene represents a spatiotemporal node. The initial population is generated by randomly selecting feasible paths in the spatiotemporal network graph. The crossover operation adopts partial mapping crossover (PMX) to ensure that the generated offspring are still valid paths. The mutation operation includes randomly replacing some spatiotemporal nodes or adjusting the time to reach a certain node. The fitness function takes multiple objectives into consideration and uses the weighted sum method or Pareto dominance relationship to evaluate the quality of individuals. The algorithm iterates the selection, crossover and mutation operations to gradually improve the quality of the population. In order to maintain the diversity of solutions, the elite retention strategy and crowding calculation are adopted. After the algorithm converges, a set of non-dominated solutions is obtained, namely the Pareto optimal path set. Each path in this set represents a balance between different objectives, providing a variety of options for the final path selection.

[0036] Specifically, selecting the final optimized path solution from the Pareto optimal path set requires comprehensive consideration of many factors. First, the system extracts the latest environmental status data from the comprehensive data set, including weather conditions, air quality, road maintenance information, etc. At the same time, the system also analyzes truck operation data, such as current location, remaining fuel, driver working hours, etc. This information will be used to evaluate the actual feasibility and adaptability of each Pareto optimal path. The evaluation process adopts multi-criteria decision analysis methods, such as the analytic hierarchy process (AHP) or the TOPSIS method. The system first defines the evaluation index system, including path length, estimated driving time, energy consumption, environmental impact, driving difficulty, etc. Then, a weight is assigned to each indicator based on the current environmental conditions and truck status. For example, the weight of the safety indicator will increase under severe weather conditions. Next, the system scores each path in the Pareto optimal path set. When calculating energy consumption, the road slope, vehicle load and current weather conditions are considered. When evaluating environmental impact, air quality data and low-emission area information are combined. When considering driving difficulty, the complexity of the path, the number of turns and the driver's fatigue are analyzed. The system also needs to check whether each path is consistent with the real-time traffic information, such as whether it passes through a road section under construction. After completing the scoring of all paths, the system will sort the paths according to the comprehensive score. The highest-ranked path is usually selected as the final optimized path plan. However, the system will also perform a final feasibility check to ensure that the selected path fully meets all hard constraints. If the highest-scoring path fails this check, the system will consider the next candidate path. The final selected optimized path plan is not only optimal in theory, but also fully adapts to the current environmental conditions and vehicle operation conditions, providing the most suitable driving plan for autonomous driving trucks.

[0037] Furthermore, according to the time window requirements in the task allocation scheme, the feasible path set is expanded in time dimension, a space-time network diagram is constructed, and a multi-objective optimization method based on a genetic algorithm is applied to the space-time network diagram to obtain a Pareto optimal path set, including: discretizing time sampling of each path in the feasible path set according to the time window requirements in the task allocation scheme to obtain a time-expanded path node set; using a dynamic time warping algorithm to align and interpolate the time-expanded path node set to obtain a regularized space-time path; constructing a directed acyclic graph based on the space-time path, and applying a topological sorting algorithm to generate a space-time network diagram that satisfies timing constraints; initializing a genetic algorithm population according to a preset multi-objective fitness function and the space-time network diagram, and performing crossover, mutation and selection operations on the genetic algorithm population, and iteratively optimizing to obtain a non-dominated solution set; using the NSGA-II algorithm to sort and screen the non-dominated solution set to obtain a Pareto optimal path set.

[0038] Specifically, the time window information of each task, including the earliest start time and the latest completion time, needs to be extracted from the task allocation plan. Then, the time dimension is expanded for each path in the feasible path set. This process starts from the starting point of the path and moves step by step along the path, creating multiple time points at each spatial node. The selection of time points is based on a predefined time interval, usually 5 minutes or 10 minutes, depending on the required accuracy and computing resources. For each spatial node on the path, the system creates all time points ranging from the earliest possible arrival time to the latest allowed arrival time. This process needs to consider the travel time of the road segment, which is calculated based on the road type, length and estimated speed. When creating time points, the system also needs to consider the constraints of the task time window to ensure that the generated time points fall within the allowed range. For paths with multiple tasks, the system needs to re-evaluate the time window at each task point, which may cause the distribution of time points at subsequent nodes to change. In addition, the system needs to consider the rest time and refueling time of the vehicle and insert additional time points at appropriate locations. The result of this process is a time-expanded path node set, in which each original spatial node is expanded into multiple spatiotemporal nodes.

[0039] Specifically, the application of the dynamic time warping (DTW) algorithm aims to solve the inconsistency problem that may exist in the time dimension of the path nodes after time expansion. First, the system selects a reference path, usually the path with the largest time span or the largest number of nodes. Then, all other paths are time-aligned with this reference path. The DTW algorithm achieves this goal by finding the best match between two paths. The algorithm starts from the starting point of the two paths and moves forward step by step, calculating the distance between the current node pair at each step (usually using Euclidean distance or Manhattan distance). Then, the algorithm selects the path that minimizes the cumulative distance to pair the nodes. This process may result in multiple nodes of one path corresponding to a single node of another path, or vice versa. To handle this situation, the system needs to perform interpolation processing. For time points where corresponding nodes are missing, the system uses linear interpolation or more complex interpolation methods (such as cubic spline interpolation) to estimate the location and attributes of the nodes. This process needs to take into account the actual shape and attributes of the road to ensure that the interpolated nodes still fall on a valid path. At the same time, the system also needs to handle time window constraints to ensure that the interpolated nodes do not violate the time requirements of the task. When interpolating, the system also needs to take into account the vehicle's motion characteristics, such as acceleration and deceleration limits, to ensure that the generated space-time path is physically feasible. The result of this process is a set of regularized space-time paths that are aligned in time and have evenly distributed nodes.

[0040] Specifically, the regularized space-time path is converted into a directed acyclic graph (DAG) to better represent and process space-time relationships. The construction process starts from the starting point, and each space-time node is regarded as a vertex in the graph. Then, the system adds directed edges between adjacent nodes. The direction of the edge represents the flow of time, and the weight of the edge can represent the travel time, distance or other costs. When adding edges, the system needs to consider the continuity of time and space to ensure that the edges only connect nodes that are adjacent in time and space. For paths containing multiple tasks, the system needs to add special marks at the task points for subsequent processing. After building the basic graph structure, the system applies a topological sorting algorithm to ensure that the graph meets the timing constraints. The topological sorting algorithm first identifies all the source points in the graph (nodes with in-degree 0), which are usually the starting points of the path. Then, the algorithm gradually removes these source points and adds them to the sorting result. Each time a node is removed, the in-degree of the nodes connected to it is updated. This process is repeated until all nodes are processed. If a node with in-degree 0 cannot be found in a certain step, it means that there is a loop in the graph, which violates the timing constraints. In this case, the system needs to backtrack and adjust the graph structure. The result of topological sorting ensures that each path in the graph is ordered in time. During the sorting process, the system also needs to verify whether each node meets the time window constraints of the task. If a violation of the constraint is found, adjustments need to be made, which may involve replanning part of the path.

[0041] Specifically, initializing the genetic algorithm population is the starting point of the optimization process. First, the system needs to define the chromosome encoding method, usually using path encoding, and each gene represents a node in the spatiotemporal network graph. The population size is determined according to the complexity of the problem and the computing resources, usually between 100 and 500. The initial population is generated by randomly selecting paths that meet the constraints on the spatiotemporal network graph. Then, the system defines a multi-objective fitness function, and typical objectives include minimizing the total driving time, energy consumption, and cost, as well as maximizing punctuality and service quality. Each goal has its own specific calculation method, such as driving time is calculated based on the length of the road section and the estimated speed, and energy consumption takes into account vehicle characteristics and road slope. Next, it enters the iterative optimization stage. In each generation, the system first performs a selection operation, using methods such as tournament selection or roulette selection. Then a crossover operation is performed, using methods such as partial mapping crossover (PMX) or sequential crossover (OX) to ensure that the generated offspring is still a valid path. The mutation operation includes randomly replacing some genes or adjusting the order of node access, and the mutation rate is usually set between 1% and 5%. After each operation, the system needs to check whether the newly generated individuals meet the constraints of the spatiotemporal network graph. When evaluating individual fitness, the system calculates the value of each objective function and uses the Pareto dominance relationship to compare the pros and cons between individuals. Non-dominated sorting divides the population into multiple fronts, and the solutions in each front are non-dominated solutions at that level. The system also uses crowding calculation to maintain the diversity of solutions and ensure that the search does not converge to the local optimum too early. The iterative process continues until a preset termination condition is reached, such as the maximum number of generations or the stability of the solution. In the end, the system obtains a set of non-dominated solutions.

[0042] Specifically, the application of NSGA-II (Non-dominated Sorting Genetic Algorithm II) aims to filter out the true Pareto optimal solutions from the non-dominated solution set. First, the system performs non-dominated sorting on the entire non-dominated solution set. This process divides the solution set into multiple frontiers, where the first front contains the non-dominated solutions in the entire population, the second front contains the non-dominated solutions in the population after removing the first frontier, and so on. During the sorting process, the system compares the advantages and disadvantages of each pair of solutions on all objectives, and calculates the domination count (how many solutions are dominated by it) and domination set (which solutions are dominated) of each solution. The time complexity of this process is O(MN^2), where M is the number of objectives and N is the population size. Next, the system calculates the crowding distance of each solution. The crowding distance measures the distribution of solutions in the objective space and is calculated by finding the normalized distance between two adjacent solutions in each objective dimension. For the boundary solutions of the frontier, infinite crowding distance is assigned to ensure that they are retained. Then, the system sorts the solutions according to the level of non-dominated sorting and crowding distance. Solutions with lower non-dominated levels are preferred, and solutions with larger crowding distances are preferred within the same level. This approach ensures the quality of the solution while maintaining the diversity of the solution set. The system also needs to set a maximum number of retained solutions, which is usually determined based on the complexity of the problem and the decision requirements. Solutions exceeding this number will be eliminated. During the screening process, the system also needs to check whether the retained solutions cover all important target trade-offs. If it is found that there are too few solutions under certain target combinations, it may be necessary to adjust the screening strategy or return to the optimization stage to guide the search to explore in these directions. The final Pareto optimal path set represents the optimal solution set that cannot be further improved while considering all goals.

[0043] 104. Based on the task allocation plan, optimized path plan, comprehensive data set and real-time status data of trucks from autonomous vehicles, predictive analysis models are used to evaluate and respond to potential risks in the transportation process, generate and execute real-time scheduling instructions.

[0044] In one embodiment of the present invention, the predictive analysis model is used to evaluate and respond to potential risks in the transportation process according to the task allocation plan, optimized path plan, comprehensive data set and real-time status data of the truck of the autonomous driving vehicle, and the generation and execution of real-time scheduling instructions include: extracting and fusing features of the comprehensive data set and the real-time status data of the truck to obtain a multi-dimensional feature vector, and inputting the multi-dimensional feature vector into a pre-trained time series prediction model to obtain a prediction result set; using a multi-criteria decision analysis method based on the prediction result set to quantitatively evaluate potential risks and generate a risk score, and using a heuristic algorithm to dynamically adjust the task allocation plan and the optimized path plan according to the risk score to obtain an adjusted scheduling strategy; converting the adjusted scheduling strategy into specific vehicle instructions and task instructions to obtain a real-time scheduling instruction set, distributing the real-time scheduling instruction set to the corresponding autonomous driving vehicle, and monitoring the execution of the real-time scheduling instruction set to achieve real-time scheduling control.

[0045] Specifically, first, feature extraction is required for the comprehensive data set and the real-time status data of the truck. From the comprehensive data set, key features related to the transportation task are extracted, such as road conditions, weather conditions, traffic flow, etc. For the real-time status data of the truck, information including location, speed, fuel level, engine status, etc. is extracted. The feature extraction process uses dimensionality reduction techniques such as principal component analysis (PCA) or autoencoders to reduce the data dimension and retain the most important information. Next, the extracted features are fused, and data fusion algorithms such as Kalman filters or particle filters are used to integrate data from different sources into a unified multi-dimensional feature vector. This feature vector contains comprehensive information about the current transportation environment and vehicle status. Then, this multi-dimensional feature vector is input into a pre-trained time series prediction model. The model can be a long short-term memory network (LSTM) or a gated recurrent unit (GRU), which can effectively capture the long-term dependencies of time series data. The model is trained using historical data, including past transportation records, weather changes, traffic patterns, etc. The prediction model outputs a prediction result set containing the predicted values ​​of various indicators in the future period, such as estimated arrival time, road condition changes, energy consumption, etc.

[0046] Specifically, based on the prediction result set, the system uses a multi-criteria decision analysis method to quantitatively evaluate potential risks. First, the evaluation indicators are defined, including delay risk, safety risk, energy consumption risk, etc. Then, the analytic hierarchy process (AHP) is used to determine the weight of each indicator, considering the relative importance of the indicators in different situations. Next, the fuzzy comprehensive evaluation method is applied to convert the prediction results into fuzzy sets, and the membership of each risk is obtained through fuzzy reasoning. Combining these memberships and weights, the overall risk score is calculated. This score reflects the overall risk level faced by the current scheduling plan. According to the generated risk score, the system uses a heuristic algorithm to dynamically adjust the task allocation plan and the optimized path plan. Using simulated annealing algorithm or genetic algorithm, with risk minimization as the objective function, possible adjustment plans are explored under the premise of ensuring task completion. The adjustment process considers multiple factors, such as task reallocation, path replanning, time window adjustment, etc. The algorithm generates candidate solutions in each iteration, evaluates their risk scores, and accepts or rejects new solutions based on a certain probability. This process continues until a solution with a significantly reduced risk score is found or the preset number of iterations is reached.

[0047] Specifically, converting the adjusted scheduling strategy into specific instructions is a key step in achieving real-time scheduling control. First, the system decomposes the scheduling strategy into specific instructions for each autonomous vehicle. These instructions include path navigation instructions, speed control instructions, task execution instructions, etc. Path navigation instructions need to consider the real-time status of the road network and convert the optimized path into a series of navigation points and turn instructions. Speed ​​control instructions provide dynamic speed recommendations for vehicles based on predicted traffic flow and road conditions. Task execution instructions detail the operating requirements and schedules for each loading and unloading point. The system integrates these instructions into a structured real-time scheduling instruction set, using standardized data formats such as JSON or XML to ensure the clarity and consistency of the instructions. Then, the instruction set is distributed to the corresponding autonomous vehicles through a secure communication protocol. The distribution process uses encrypted transmission to ensure the security and integrity of the instructions. At the same time, the system establishes a real-time monitoring mechanism to continuously track the location, status and instruction execution progress of each vehicle. The monitoring system uses real-time data stream processing technology to quickly process large amounts of sensor data and status reports. When instruction execution deviations or new risk factors are detected, the system triggers an immediate adjustment mechanism to generate corrective instructions or re-evaluate risk and optimize scheduling.

[0048] Furthermore, the use of a multi-criteria decision analysis method to quantitatively evaluate potential risks based on the prediction result set, generate a risk score, and use a heuristic algorithm to dynamically adjust the task allocation plan and the optimization path plan according to the risk score to obtain an adjusted scheduling strategy, including: constructing a decision matrix based on the prediction result set, and using a multi-criteria decision analysis method and a decision matrix to calculate the score of each potential risk to obtain a risk score set; determining the tasks and paths that need to be adjusted based on the risk score set, and generating an adjustment target list; using a heuristic algorithm to reallocate tasks and adjust paths in the adjustment target list to generate multiple candidate scheduling plans; evaluating the risk level and scheduling efficiency of the candidate scheduling plans, and selecting the optimal plan as the adjusted scheduling strategy.

[0049] Specifically, the rows of the decision matrix represent different potential risk factors, such as delay risk, safety risk, energy consumption risk, etc., and the columns represent different evaluation indicators, such as probability of occurrence, degree of impact, duration, etc. Each element of the matrix represents the predicted value of a specific risk factor under a specific indicator. After the construction is completed, the system uses a multi-criteria decision analysis method, such as the analytic hierarchy process (AHP) or the TOPSIS method, to process the decision matrix. In the AHP method, a hierarchical structure is first established to decompose the risk assessment problem into three levels: goals, criteria, and solutions. Then, through expert judgment or historical data analysis, a judgment matrix is ​​constructed to calculate the weight of each criterion. Next, a consistency test is used to ensure the rationality of the judgment. Finally, the weights of each level are combined to obtain a comprehensive score for each risk factor. If the TOPSIS method is used, it is necessary to determine the positive ideal solution and the negative ideal solution, calculate the distance from each risk solution to the ideal solution, and sort them accordingly. Through this process, the system obtains a risk score set that contains a quantitative score for each potential risk.

[0050] Specifically, based on the risk score set obtained, the system needs to determine which tasks and paths need to be adjusted. First, a risk threshold is set, which can be determined through historical data analysis or expert experience. Then, each risk score is compared with the threshold to identify high-risk items that exceed the threshold. For these high-risk items, the system analyzes their impact range and determines the specific tasks and paths affected. This process needs to consider the propagation effect of risk, that is, a high-risk factor may have a chain reaction on multiple related tasks or paths. Next, the system prioritizes the tasks and paths that need to be adjusted. The sorting criteria include factors such as risk level, task urgency, and adjustment difficulty. Projects with high risk, high urgency, and low adjustment difficulty will be given priority. The system also needs to evaluate the feasibility of the adjustment, considering constraints such as resource constraints, time windows, and vehicle capabilities. Based on these analyses, the system generates a list of adjustment targets. This list contains detailed information about the tasks and paths that need to be adjusted, such as task ID, currently assigned vehicle, original path, risk type and degree, etc. At the same time, the list also contains the priority of each adjustment target and preliminary adjustment suggestions.

[0051] Specifically, for the adjustment target list, the system uses a heuristic algorithm to redistribute tasks and adjust paths. The reason for choosing a heuristic algorithm is that it can find a solution close to the optimal in a short time, which is suitable for the needs of real-time scheduling. Specifically, a simulated annealing algorithm or a genetic algorithm can be used. Taking the simulated annealing algorithm as an example, the representation of the solution is first defined. Usually, a task-vehicle assignment matrix and a path sequence are used to represent a complete scheduling plan. The initial solution is based on the current scheduling plan, and then a new candidate solution is generated in each iteration. The method of generating a new solution includes randomly exchanging the assignment of two tasks, adjusting the order of nodes in the path, inserting or deleting nodes in the path, etc. For each newly generated solution, its objective function value is calculated. The objective function needs to comprehensively consider factors such as the degree of risk reduction, scheduling efficiency, and resource utilization. If the new solution is better than the current solution, the new solution is accepted; if the new solution is worse, it is accepted with a certain probability. This probability decreases with the increase in the number of iterations, simulating the physical annealing process. The algorithm maintains an optimal solution set and records the best solutions encountered during the iteration process. In order to improve efficiency, a local search strategy can be introduced, such as performing a small-scale optimization after each major adjustment. At the same time, the algorithm needs to consider various constraints, such as time windows, vehicle capacity, driving time limits, etc., to ensure that the generated plan is feasible. After a preset number of iterations or reaching convergence conditions, the algorithm outputs multiple candidate scheduling plans, which all adjust the original high-risk tasks and paths, but may have different trade-offs in different aspects.

[0052] Specifically, after obtaining multiple candidate scheduling schemes, the system needs to conduct a comprehensive evaluation of these schemes to select the best one. The evaluation process first needs to define an evaluation index system, including risk indicators (such as overall risk score, proportion of high-risk tasks) and efficiency indicators (such as total driving distance, task completion time, and vehicle utilization). For each candidate scheme, the system uses the previously established risk assessment model to recalculate the risk level, and uses a simulation model or a mathematical model to evaluate the scheduling efficiency. The evaluation process adopts a multi-objective evaluation method, and the evaluation results will form a comprehensive scoring table, including the scores of each candidate scheme on each indicator and the overall ranking. The system selects the scheme with the highest comprehensive score as the final adjusted scheduling strategy. If there are multiple schemes with similar scores, further consideration can be given to factors such as implementation difficulty and similarity with the original scheme to make decisions. Finally, the system will generate a detailed scheme description, including the specific content of the adjustment, expected effects, potential risks, etc., to provide guidance for actual implementation.

[0053] In this embodiment, a comprehensive data set is generated by multi-dimensional collection and preprocessing of the autonomous driving vehicle. Based on the comprehensive data set, the loading and unloading tasks are dynamically allocated and the fleet is collaboratively planned to generate a task allocation plan. Through a hierarchical search algorithm and a multi-objective optimization algorithm, the vehicle's path is planned and optimized in combination with the task allocation plan and the comprehensive data set to generate an optimized path plan. Based on the task allocation plan, the optimized path plan, the comprehensive data set and the real-time status data of the vehicle, a predictive analysis model is used to evaluate and respond to potential risks in the transportation process, and real-time scheduling instructions are generated and executed. The present invention combines multi-dimensional data collection and processing, dynamic task allocation, path optimization and risk prediction scheduling methods to comprehensively improve the scheduling efficiency and safety of autonomous driving vehicles and achieve efficient logistics management.

[0054] The above describes the dispatching method of the autonomous driving vehicle in the embodiment of the present invention. The following describes the dispatching device of the autonomous driving vehicle in the embodiment of the present invention. Figure 2 , an embodiment of the dispatching device of the autonomous driving vehicle in the embodiment of the present invention includes: The data processing module 201 is used to collect and preprocess the truck operation data, road information data, cargo characteristic data and environmental condition data of the autonomous driving vehicle in multiple dimensions to obtain a preprocessed comprehensive data set; The task allocation module 202 is used to dynamically allocate the preset loading and unloading tasks and coordinate fleet planning based on the comprehensive data set using the preset cargo classification model and the rolling time domain optimization strategy to obtain a task allocation plan; A path optimization module 203 is used to plan and optimize the truck path of the autonomous driving vehicle according to the task allocation scheme and the comprehensive data set through a preset hierarchical search algorithm and a multi-objective optimization algorithm to obtain an optimized path plan; The scheduling module 204 is used to evaluate and respond to potential risks in the transportation process using a predictive analysis model based on the task allocation plan, optimized path plan, comprehensive data set and real-time status data of the truck of the autonomous driving vehicle, and generate and execute real-time scheduling instructions.

[0055] In an embodiment of the present invention, the dispatching device of the autonomous driving vehicle runs the dispatching method of the autonomous driving vehicle described above, and the dispatching device of the autonomous driving vehicle generates a comprehensive data set by multi-dimensional collection and preprocessing of the autonomous driving vehicle. Based on the comprehensive data set, the loading and unloading tasks are dynamically allocated and the fleet is collaboratively planned to generate a task allocation plan. Through a hierarchical search algorithm and a multi-objective optimization algorithm, the vehicle's path is planned and optimized in combination with the task allocation plan and the comprehensive data set, and an optimized path plan is generated. According to the task allocation plan, the optimized path plan, the comprehensive data set and the real-time status data of the vehicle, a predictive analysis model is used to evaluate and respond to potential risks in the transportation process, and real-time scheduling instructions are generated and executed. The present invention combines a scheduling method of multi-dimensional data collection and processing, dynamic task allocation, path optimization and risk prediction to comprehensively improve the scheduling efficiency and safety of autonomous driving vehicles and achieve efficient logistics management.

[0056] above Figure 2 The dispatching device of the autonomous driving vehicle in the embodiment of the present invention is described in detail from the perspective of modular functional entities. The dispatching device of the autonomous driving vehicle in the embodiment of the present invention is described in detail from the perspective of hardware processing.

[0057] Figure 33 is a schematic diagram of the structure of a dispatching device for an autonomous driving vehicle provided by an embodiment of the present invention. The dispatching device 300 for the autonomous driving vehicle may have relatively large differences due to different configurations or performances, and may include one or more processors (central processing units, CPU) 310 (for example, one or more processors) and a memory 320, and one or more storage media 330 (for example, one or more mass storage device terminals) storing application programs 333 or data 332. Among them, the memory 320 and the storage medium 330 may be short-term storage or persistent storage. The program stored in the storage medium 330 may include one or more modules (not shown in the figure), and each module may include a series of instruction operations in the dispatching device 300 for the autonomous driving vehicle. Furthermore, the processor 310 may be configured to communicate with the storage medium 330, and execute a series of instruction operations in the storage medium 330 on the dispatching device 300 for the autonomous driving vehicle to implement the steps of the dispatching method for the autonomous driving vehicle described above.

[0058] The dispatching device 300 of the autonomous driving vehicle may also include one or more power supplies 340, one or more wired or wireless network interfaces 350, one or more input and output interfaces 360, and / or one or more operating systems 331, such as Windows Serve, Mac OS X, Unix, Linux, FreeBSD, etc. It will be appreciated by those skilled in the art that Figure 3 The structure of the dispatching device for the autonomous driving vehicle shown does not constitute a limitation on the dispatching device for the autonomous driving vehicle provided by the present invention, and may include more or fewer components than shown in the figure, or a combination of certain components, or a different arrangement of components.

[0059] The present invention also provides a computer-readable storage medium, which may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. Instructions are stored in the computer-readable storage medium. When the instructions are executed on a computer, the computer executes the steps of the scheduling method for the autonomous driving vehicle.

[0060] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the above-described system, device, or unit can refer to the corresponding process in the aforementioned method embodiment and will not be repeated here.

[0061] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art or the whole or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk and other media that can store program code.

[0062] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features thereof may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for dispatching an autonomous driving vehicle, characterized in that: The dispatching method of the autonomous driving vehicle includes: The truck operation data, road information data, cargo characteristics data and environmental condition data of the autonomous driving vehicle are collected and preprocessed in multiple dimensions to obtain a preprocessed comprehensive data set; Using a preset cargo classification model and a rolling horizon optimization strategy, based on the comprehensive data set, dynamically allocate preset loading and unloading tasks and coordinate fleet planning to obtain a task allocation plan; By using a preset hierarchical search algorithm and a multi-objective optimization algorithm, a truck path of the autonomous driving vehicle is planned and optimized according to the task allocation scheme and the comprehensive data set to obtain an optimized path plan; Based on the task allocation plan, optimized path plan, comprehensive data set and real-time status data of the truck of the autonomous driving vehicle, a predictive analysis model is used to evaluate and respond to potential risks in the transportation process, and real-time scheduling instructions are generated and executed.

2. The method for dispatching an autonomous driving vehicle according to claim 1, characterized in that: The preset cargo classification model and rolling time domain optimization strategy are used to dynamically allocate preset loading and unloading tasks and coordinate fleet planning based on the comprehensive data set, and the task allocation scheme obtained includes: Performing cluster analysis on the cargo characteristic data in the comprehensive data set to obtain a cargo category matrix, and using an improved Hungarian algorithm to perform preliminary cargo-vehicle matching according to the cargo category matrix and the load capacity of the autonomous driving vehicle to obtain an initial allocation plan; Applying a task merging algorithm based on time and space constraints to the initial allocation plan, integrating the loading and unloading tasks that meet the conditions to obtain an optimized task combination; Using a dynamic programming algorithm, according to the optimized task combination and the truck operation data, the optimal loading and unloading sequence is calculated to obtain a preliminary scheduling sequence; The preliminary scheduling sequence is dynamically updated within a preset time window by using a rolling time domain optimization strategy, and new tasks are inserted into the preliminary scheduling sequence in real time to obtain a task allocation plan.

3. The method for dispatching an autonomous driving vehicle according to claim 2, characterized in that: The task merging algorithm based on time and space constraints is applied to the initial allocation plan to integrate the qualified loading and unloading tasks to obtain the optimized task combination including: Performing time-space coordinate mapping on the loading and unloading tasks in the initial allocation plan to obtain a task time-space distribution matrix; Using a density clustering algorithm to perform cluster analysis on the task spatiotemporal distribution matrix, a potential mergeable task set is obtained; According to a preset time window and space distance threshold, the potentially merging task set is screened to obtain task pairs that meet the merging conditions; The maximum weight matching algorithm in graph theory is used to optimally combine the task pairs that meet the merging conditions to obtain a preliminary merging solution; A heuristic search algorithm is applied to the preliminary merging scheme, and local optimization adjustment is performed considering vehicle capacity and path constraints to obtain an optimized task combination.

4. The method for dispatching an autonomous driving vehicle according to claim 1, characterized in that: The preset hierarchical search algorithm and multi-objective optimization algorithm are used to plan and optimize the truck path of the autonomous driving vehicle according to the task allocation scheme and the comprehensive data set, and the optimized path scheme includes: According to the road information data in the comprehensive data set, a multi-level road network model is constructed, and a coarse-grained path search is performed in the multi-level road network model using an improved Dijkstra algorithm to obtain a candidate path set, wherein the multi-level road network model is a hierarchical graph structure including trunk roads and secondary roads; Applying the A* algorithm that takes into account the turning radius and height restrictions of the truck to the candidate path set, performing fine-grained path search, and obtaining a feasible path set that meets the constraint conditions; According to the time window requirement in the task allocation scheme, the feasible path set is expanded in time dimension, a space-time network diagram is constructed, and a multi-objective optimization method based on a genetic algorithm is applied to the space-time network diagram to obtain a Pareto optimal path set; According to the environmental condition data and the truck operation data in the comprehensive data set, the most suitable path is selected from the Pareto optimal path set as the optimized path solution.

5. The method for dispatching an autonomous driving vehicle according to claim 4, characterized in that: According to the time window requirement in the task allocation scheme, the feasible path set is expanded in time dimension, a space-time network diagram is constructed, and a multi-objective optimization method based on a genetic algorithm is applied to the space-time network diagram to obtain a Pareto optimal path set including: Discrete time sampling is performed on each path in the feasible path set according to the time window requirement in the task allocation scheme to obtain a time-expanded path node set; Using a dynamic time warping algorithm, aligning and interpolating the time-expanded path node set to obtain a regularized space-time path; Constructing a directed acyclic graph based on the space-time path, and applying a topological sorting algorithm to generate a space-time network graph that satisfies the timing constraints; Initializing a genetic algorithm population according to a preset multi-objective fitness function and the spatiotemporal network graph, and performing crossover, mutation and selection operations on the genetic algorithm population, and iteratively optimizing to obtain a non-dominated solution set; The non-dominated solution set is sorted and screened using the NSGA-II algorithm to obtain a Pareto optimal path set.

6. The method for dispatching an autonomous driving vehicle according to claim 1, characterized in that: The method of using a predictive analysis model to evaluate and respond to potential risks in the transportation process based on the task allocation plan, the optimized path plan, the comprehensive data set and the real-time status data of the truck of the autonomous driving vehicle, and generating and executing real-time dispatch instructions includes: Extracting and fusing features of the comprehensive data set and the real-time status data of the truck to obtain a multi-dimensional feature vector, and inputting the multi-dimensional feature vector into a pre-trained time series prediction model to obtain a prediction result set; Using a multi-criteria decision analysis method to quantitatively evaluate potential risks based on the prediction result set, generate a risk score, and use a heuristic algorithm to dynamically adjust the task allocation plan and the optimized path plan according to the risk score to obtain an adjusted scheduling strategy; The adjusted scheduling strategy is converted into specific vehicle instructions and task instructions to obtain a real-time scheduling instruction set, the real-time scheduling instruction set is distributed to the corresponding autonomous driving vehicle, and the execution of the real-time scheduling instruction set is monitored to achieve real-time scheduling control.

7. The method for dispatching an autonomous driving vehicle according to claim 6, characterized in that: The multi-criteria decision analysis method is used to quantitatively evaluate potential risks based on the prediction result set, generate risk scores, and dynamically adjust the task allocation plan and the optimized path plan according to the risk scores using a heuristic algorithm. The adjusted scheduling strategy includes: Constructing a decision matrix based on the prediction result set, and calculating the score of each potential risk using a multi-criteria decision analysis method and the decision matrix to obtain a risk score set; Determine the tasks and paths that need to be adjusted according to the risk score set, and generate an adjustment target list; Using a heuristic algorithm to reallocate and adjust the paths of the tasks in the adjustment target list to generate multiple candidate scheduling solutions; The risk level and scheduling efficiency of the candidate scheduling solutions are evaluated, and the optimal solution is selected as the adjusted scheduling strategy.

8. A dispatching device for an autonomous driving vehicle, characterized in that: The dispatching device of the autonomous driving vehicle comprises: A data processing module is used to collect and preprocess the truck operation data, road information data, cargo characteristics data and environmental condition data of the autonomous driving vehicle in multiple dimensions to obtain a preprocessed comprehensive data set; A task allocation module, which is used to dynamically allocate preset loading and unloading tasks and coordinate fleet planning based on the comprehensive data set by using a preset cargo classification model and a rolling time domain optimization strategy to obtain a task allocation plan; A path optimization module, for planning and optimizing the truck path of the autonomous driving vehicle according to the task allocation scheme and the comprehensive data set through a preset hierarchical search algorithm and a multi-objective optimization algorithm to obtain an optimized path plan; The scheduling module is used to evaluate and respond to potential risks in the transportation process using a predictive analysis model based on the task allocation plan, optimized path plan, comprehensive data set and real-time status data of the truck of the autonomous driving vehicle, and generate and execute real-time scheduling instructions.

9. A dispatching device for an autonomous driving vehicle, characterized in that: The dispatching device of the autonomous driving vehicle includes: a memory and at least one processor, wherein the memory stores instructions; The at least one processor calls the instructions in the memory so that the scheduling device of the autonomous driving vehicle executes the steps of the scheduling method of the autonomous driving vehicle as described in any one of claims 1-7.

10. A computer-readable storage medium having instructions stored thereon, characterized in that: When the instructions are executed by the processor, the steps of the scheduling method for the autonomous driving vehicle as described in any one of claims 1-7 are implemented.

Images