Path planning and control method and system for container port automatic driving vehicle

Abstract

The invention relates to the technical field of path planning, and discloses a path planning and control method and system for a container port automatic driving vehicle. The method comprises the following steps: acquiring environment data through a multi-source sensor, and obtaining a harbor district digital twinborn model through space-time registration fusion; a three-layer progressive planner is used for optimizing and generating a main and standby path scheme based on the model; predicting a dynamic target trajectory by using a time sequence attention network to obtain situation assessment; processing the evaluation data through a distributed predictive control algorithm to generate a control instruction; and monitoring a motion state through adaptive security constraints to form a security policy, and continuously optimizing by using deep reinforcement learning. According to the method, safe and reliable operation of the automatic driving vehicle is realized in a complex and changeable harbor district environment, and particularly, accurate path planning and control decisions can be made in the face of dynamic obstacles.

Description

Technical Field

[0001] The present application relates to the field of path planning technology, and in particular to a path planning and control method and system for an autonomous driving vehicle in a container port. Background Art

[0002] Currently, fixed path planning and simple obstacle avoidance mechanisms are mainly used in the navigation and control systems of autonomous driving vehicles in container ports. Such systems usually rely on preset path points and basic sensor data, and use a single path planning algorithm for navigation. At the same time, existing port autonomous driving systems mostly use independent perception, planning and control modules, with limited information exchange between modules, and relatively fixed control strategies, lacking the ability to adapt to complex dynamic environments.

[0003] However, this traditional technical solution has obvious shortcomings: first, fixed path planning cannot effectively cope with the frequently changing dynamic obstacles in the port area; second, the simple modular design makes it impossible for the system to fully utilize multi-source sensor data for collaborative decision-making; third, the lack of adaptive safety control mechanism can easily lead to safety hazards in complex scenarios; finally, existing systems generally lack the ability to learn and optimize historical data, making it difficult to achieve continuous improvement of control strategies. Summary of the invention

[0004] The present application provides a path planning and control method and system for an autonomous container port vehicle, which is used to achieve safe and reliable operation of the autonomous vehicle in a complex and changeable port environment, especially when facing dynamic obstacles, and can make accurate path planning and control decisions.

[0005] In the first aspect, the present application provides a path planning and control method for an autonomous driving vehicle in a container port, and the path planning and control method for an autonomous driving vehicle in a container port includes: collecting port area environmental data through a multi-source sensor network, and fusing the environmental data through a spatiotemporal registration algorithm to obtain a digital twin environment model of the port area; according to the digital twin environment model of the port area, hierarchically optimizing the vehicle path through a three-layer progressive path planner to obtain a path planning scheme including a main path and an alternative path; based on the path planning scheme, using a temporal attention network to perform trajectory prediction and analysis on dynamic targets in the port area to obtain real-time situation assessment data; based on the real-time situation assessment data, using a distributed model predictive control algorithm to perform collaborative control processing on the vehicle to generate a vehicle control instruction sequence; according to the vehicle control instruction sequence, using an adaptive safety constraint module to monitor the vehicle's motion state in real time to form a multi-level safety response strategy, and using a deep reinforcement learning network to continuously optimize the control strategy.

[0006] In a second aspect, the present application provides a path planning and control system for an autonomous container port vehicle, the path planning and control system for an autonomous container port vehicle comprising: A fusion module is used to collect port area environmental data through a multi-source sensor network, and to fuse the environmental data through a spatiotemporal registration algorithm to obtain a digital twin environmental model of the port area; A hierarchical module, used to perform hierarchical optimization processing on the vehicle path through a three-layer progressive path planner according to the port area digital twin environment model, and obtain a path planning solution including a main path and an alternative path; A prediction module, which is used to perform trajectory prediction analysis on dynamic targets in the port area based on the path planning scheme using a temporal attention network to obtain real-time situation assessment data; A control module, configured to perform coordinated control processing on the vehicle through a distributed model predictive control algorithm based on the real-time situation assessment data to generate a vehicle control instruction sequence; The monitoring module is used to monitor the vehicle's motion state in real time through the adaptive safety constraint module according to the vehicle control instruction sequence, form a multi-level safety response strategy, and continuously optimize the control strategy using a deep reinforcement learning network.

[0007] A third aspect of the present invention provides a computer device, comprising: 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 computer device executes the above-mentioned path planning and control method for the container port autonomous driving vehicle.

[0008] 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 above-mentioned path planning and control method for an autonomous driving vehicle in a container port.

[0009] In the technical solution provided by this application, the port area environmental data is collected through a multi-source sensor network, and the spatiotemporal registration algorithm is used for fusion processing to construct a high-precision digital twin environment model, which effectively solves the problem of incomplete and inaccurate data from traditional single sensors, and significantly improves the comprehensiveness and accuracy of environmental perception. The three-layer progressive path planner is used to perform hierarchical optimization of vehicle paths, which can not only generate the optimal main path, but also include multiple alternative paths, enhance the system's ability to respond to emergencies, and improve the robustness of path planning. The introduction of the temporal attention network enables the system to accurately predict the trajectory of dynamic targets in the port area. Through deep learning and temporal correlation analysis of target motion characteristics, the system's prediction accuracy of dynamic environmental changes is greatly improved. The application of distributed model predictive control algorithm realizes collaborative control among multiple vehicles, and can dynamically adjust the control strategy based on real-time situation assessment data, significantly improving the efficiency and safety of group control. The adaptive safety constraint module cooperates with the continuous optimization mechanism of the deep reinforcement learning network, so that the system can continuously improve the control strategy according to the actual operation data, and establish an adaptive evolutionary safety control system. Especially in terms of algorithm features, the temporal attention network extracts the key temporal features of dynamic targets through the attention mechanism processing of time series data, making trajectory prediction more accurate; the deep reinforcement learning network continuously improves the intelligence and adaptability of control decisions through continuous interaction with the environment and strategy optimization. The comprehensive application of these algorithm features significantly enhances the system's adaptability and control performance in complex port environments. Overall, the solution has built an intelligent and adaptive port autonomous driving control system through the organic combination of multiple innovative algorithms, achieving all-round optimization of environmental perception, path planning, trajectory prediction, collaborative control and safety monitoring, and greatly improving the operating efficiency and safety of port autonomous driving vehicles. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0011] Figure 1 This is a schematic diagram of an embodiment of a method for path planning and control of an autonomous driving vehicle in a container port in an embodiment of the present application; Figure 2 This is a schematic diagram of an embodiment of a path planning and control system for an autonomous driving vehicle in a container port in an embodiment of the present application; Figure 3 It is a schematic block diagram of the structure of a computer device in an embodiment of the present invention. DETAILED DESCRIPTION

[0012] Embodiments of the present application provide a path planning and control method and system for an autonomous driving vehicle in a container port. The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "including" or "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0013] For ease of understanding, the specific process of the embodiment of the present application is described below. Figure 1 , an embodiment of the path planning and control method of the container port autonomous driving vehicle in the embodiment of the present application includes: Step S101: Collect port area environmental data through a multi-source sensor network, fuse the environmental data through a spatiotemporal registration algorithm, and obtain a digital twin environment model of the port area; Step S102: According to the port area digital twin environment model, the vehicle path is hierarchically optimized by a three-layer progressive path planner to obtain a path planning solution including a main path and an alternative path; Step S103: Based on the path planning scheme, the temporal attention network is used to perform trajectory prediction analysis on the dynamic targets in the port area to obtain real-time situation assessment data; Step S104: Based on the real-time situation assessment data, the vehicle is cooperatively controlled by a distributed model predictive control algorithm to generate a vehicle control instruction sequence; Step S105: According to the vehicle control command sequence, the vehicle's motion state is monitored in real time via the adaptive safety constraint module to form a multi-level safety response strategy, and the control strategy is continuously optimized using a deep reinforcement learning network.

[0014] It is understandable that the execution subject of this application can be a path planning and control system for an autonomous driving vehicle in a container port, or a terminal or a server, which is not limited here. The embodiment of this application is described by taking a server as the execution subject as an example.

[0015] Specifically, the port area environmental data is collected through a multi-source sensor network. The multi-source sensor network includes a laser radar array, a high-definition visual sensor array, and a millimeter-wave radar network, which are responsible for collecting different types of environmental information. The laser radar array collects three-dimensional point cloud data of the port area and records the spatial position information of static objects such as containers and fixed facilities; the high-definition visual sensor array obtains real-time image data of the port area for identifying and classifying various types of targets in the port area; the millimeter-wave radar network detects the speed and direction information of moving targets. The spatiotemporal registration algorithm fuses these heterogeneous data, including three key steps: time synchronization, aligning the data collected by different sensors according to a unified timestamp; then performing spatial registration, converting the data in different coordinate systems into a unified reference coordinate system; performing data fusion, weighting and synthesizing multi-source data through a Kalman filter, and obtaining a high-precision digital twin environment model of the port area.

[0016] After obtaining the digital twin environment model of the port area, the three-layer progressive path planner performs hierarchical optimization processing on the vehicle path. The first layer is the global path planning layer, which uses the improved A* algorithm to generate a coarse-grained path plan based on the static map information of the port area; the second layer is the local path planning layer, which combines real-time dynamic information and uses the spatiotemporal RRT* algorithm to locally optimize the path, considering the impact of dynamic obstacles; the third layer is the trajectory generation layer, which generates a smooth driving trajectory through the model predictive control method. The planning results include a main path and multiple alternative paths to form a path planning plan. Based on the path planning plan, the temporal attention network performs trajectory prediction and analysis on dynamic targets in the port area. The network processes historical trajectory data through an encoder-decoder structure, and the attention mechanism focuses on the key factors that affect the target movement. The network input includes the historical position, speed and other state information of the target, and outputs the predicted future trajectory distribution. Through the multi-target trajectory association algorithm, the motion prediction results of different targets are integrated to form real-time situation assessment data.

[0017] The distributed model predictive control algorithm performs collaborative control processing on vehicles based on real-time situation assessment data. The algorithm establishes a vehicle dynamics model in the prediction time domain and transforms multi-vehicle collaborative constraints and obstacle avoidance requirements into optimization problems. Through the distributed computing architecture, multiple vehicles solve their own control problems in parallel while considering the interaction between vehicles. The optimization results generate a vehicle control command sequence, including specific control quantities such as speed and steering. The adaptive safety constraint module monitors the vehicle motion state in real time and dynamically adjusts the safety constraint boundary according to the vehicle control command sequence and current state. This module constructs a multi-level safety response strategy, including normal operation, warning, deceleration, emergency braking and other different levels. The deep reinforcement learning network learns the optimal control strategy through continuous interaction with the environment and continuously optimizes the safety response decision.

[0018] When an autonomous vehicle needs to transport a container from area A to area B in the port, the system collects environmental data through multi-source sensors. The lidar detects the location information of surrounding static obstacles, the visual sensor identifies the moving forklift and pedestrians, and the millimeter-wave radar measures their movement speed. The spatiotemporal registration algorithm fuses these data into a unified environmental representation, and the three-layer progressive path planner generates a main path and two alternative paths. During driving, the temporal attention network predicts that the forklift in front may make a turn, and the distributed model predictive control algorithm adjusts the vehicle speed and path accordingly. When potential risks are detected, the adaptive safety constraint module initiates the corresponding level of safety response to ensure the safety and efficiency of the transportation process. Data processing and decision-making throughout the process are carried out in real time, and the various modules of the system maintain close data interaction and logical association.

[0019] In the embodiment of the present application, the port area environmental data is collected through a multi-source sensor network, and the spatiotemporal registration algorithm is used for fusion processing to construct a high-precision digital twin environment model, which effectively solves the problem of incomplete and inaccurate data of traditional single sensor, and significantly improves the comprehensiveness and accuracy of environmental perception. The three-layer progressive path planner is used to hierarchically optimize the vehicle path, which can not only generate the optimal main path, but also include multiple alternative paths, enhance the system's ability to cope with emergencies, and improve the robustness of path planning. The introduction of the temporal attention network enables the system to accurately predict the trajectory of dynamic targets in the port area, and greatly improves the system's prediction accuracy of dynamic environmental changes through deep learning and temporal correlation analysis of target motion characteristics. The application of the distributed model predictive control algorithm realizes the collaborative control of multiple vehicles, and can dynamically adjust the control strategy based on real-time situation assessment data, significantly improving the efficiency and safety of group control. The adaptive safety constraint module cooperates with the continuous optimization mechanism of the deep reinforcement learning network, so that the system can continuously improve the control strategy according to the actual operation data, and establish an adaptive evolutionary safety control system. Especially in terms of algorithm features, the temporal attention network extracts the key temporal features of dynamic targets through the attention mechanism processing of time series data, making trajectory prediction more accurate; the deep reinforcement learning network continuously improves the intelligence and adaptability of control decisions through continuous interaction with the environment and strategy optimization. The comprehensive application of these algorithm features significantly enhances the system's adaptability and control performance in complex port environments. Overall, the solution has built an intelligent and adaptive port autonomous driving control system through the organic combination of multiple innovative algorithms, achieving all-round optimization of environmental perception, path planning, trajectory prediction, collaborative control and safety monitoring, and greatly improving the operating efficiency and safety of port autonomous driving vehicles.

[0020] In a specific embodiment, the process of executing step S101 may specifically include the following steps: (1) In the step of collecting port area environmental data via a multi-source sensor network, the environmental data is divided into a static environmental data stream and a dynamic environmental data stream; (2) Perform spatial coordinate analysis on the static environmental data stream and establish an index table of port infrastructure locations; (3) Perform target identification and classification on dynamic environment data streams to generate real-time target status records; (4) Pair the port infrastructure location index table with the real-time target status records by timestamp to form a time series environmental feature matrix; (5) Perform data cleaning and outlier processing on the time series environmental feature matrix to construct a standardized environmental data set; (6) The standardized environmental data set is input into the spatiotemporal registration algorithm to generate a digital twin environment model of the port area through feature mapping and data alignment.

[0021] Specifically, the environmental data collected by the multi-source sensor network needs to be processed by data stream classification. According to the data characteristics, the environmental data is divided into static environmental data stream and dynamic environmental data stream. The static environmental data stream contains the spatial position, size, type and other information of the fixed facilities of the port that do not change with time, while the dynamic environmental data stream contains the target information of mobile vehicles, operating equipment and other real-time changes. The processing of the static environmental data stream is carried out in the coordinate system 1, and the spatial data collected by different sensors are converted into a unified world coordinate system. The converted data uses the spatial clustering algorithm to identify the boundaries of different facilities and extract the geometric features of the facilities, including key parameters such as the center point coordinates, orientation angle, and occupied area. Based on these features, an index table of the location of port infrastructure is established. The index table adopts a multi-level tree structure, and each node records the location information, attribute labels and spatial relationships of the facilities.

[0022] The processing of dynamic environment data streams involves target detection and tracking. The sensor data is classified for target recognition and the detected dynamic targets are divided into different categories such as autonomous vehicles, manually operated vehicles, and operating machinery. The motion state parameters such as position, speed, acceleration, etc. are extracted for each target, and the type label, size information, and behavioral characteristics of the target are recorded. This information constitutes a real-time target state record, and each record is accompanied by an accurate timestamp. The time series matching process associates the port infrastructure location index table with the real-time target state record. The data is aligned according to the timestamp to establish a time series environment feature matrix. Each row of the matrix represents a time point, and the columns contain static environment information and dynamic target status to form a scene description. There are spatial and temporal correlations between the matrix elements, reflecting the characteristics of the port environment evolving over time.

[0023] The time series environmental feature matrix is ​​cleaned, mainly dealing with outliers and missing values. Outlier detection uses a statistical method to calculate the distribution characteristics of each feature and identify data points that deviate significantly from the normal range. For the detected outliers, they are corrected or eliminated in combination with the data before and after. Missing values ​​are supplemented by spatiotemporal interpolation methods to maintain the continuity and consistency of the data. The cleaned data is standardized, and the features of different dimensions are converted to a unified scale space to construct a standardized environmental data set. The standardized environmental data set is input into the spatiotemporal registration algorithm. The algorithm performs feature mapping, projects the environmental features into a high-dimensional feature space, and extracts the key features that characterize the environmental state. Then, data alignment is performed to establish the correspondence between different times and different areas to eliminate spatial and temporal inconsistencies. Through these processes, a digital twin environmental model of the port area is generated, which accurately reflects the real-time status of the port area.

[0024] For example, when the laser radar scans to obtain the point cloud data of the container stack in the yard, the visual camera collects images of lane markings and traffic signs, and the millimeter wave radar tracks the forklifts operating in the yard. After these raw data are classified, the static data includes the container location coordinates, stack height, channel width, etc., and the dynamic data includes the real-time position and speed of the forklift. The position index table established by spatial coordinate analysis records the precise location and occupied space of each container stack, and the target recognition classification generates the motion trajectory record of the forklift. Time series matching integrates this information into the feature matrix. Each element in the matrix has a clear physical meaning. After outlier processing and standardization, it is input into the spatiotemporal registration algorithm. The registration process eliminates the errors between different sensors. The generated digital twin model fully describes the static layout and dynamic operation status of the yard, providing reliable environmental information for the path planning of autonomous driving vehicles.

[0025] In a specific embodiment, the process of executing step S102 may specifically include the following steps: (1) The digital twin environment model of the port area is parsed into a road network topology diagram, the vehicle traffic attributes of the road network nodes are marked, and a basic road network data set is generated; (2) Extract the node distance matrix and turning constraints from the road network basic data set and construct the initial cost function for path planning; (3) Assign weights to the initial cost function of path planning according to vehicle traffic priority and generate a multi-level path evaluation standard; (4) Decomposing the multi-level path evaluation criteria through a three-layer progressive path planner to form a set of path candidates; (5) Perform collision risk assessment and travel time calculation on the candidate path set to select the main path; (6) Determine the search range of alternative paths based on the evaluation results of the main path and generate a path planning solution.

[0026] Specifically, the digital twin environment model of the port area needs to be converted into a road network topology diagram. This process uses graphics algorithms to extract the centerline of the road and identify the road network nodes and connection relationships. The traffic attributes of each road network node are marked, including node type (intersection, turning point, loading and unloading point, etc.), traffic direction, speed limit and other information. This information constitutes the basic data set of the road network, and the adjacency matrix is ​​used to store the connection relationship between nodes. The node distance matrix is ​​extracted from the basic data set of the road network to record the shortest path distance between any two nodes. At the same time, turning constraints are extracted, including rules such as prohibited turning and forced turning. Based on these data, the initial cost function of path planning is constructed, taking into account multiple factors such as path length, number of turns, and travel time. The calculation formula of the cost function is as follows: in, represents the total cost of path p, represents the distance of the i-th path, represents the turning cost of the j-th turning point, represents the speed influencing factor of the kth node, , , is the weight coefficient. The cost function is weighted according to the vehicle priority. The higher the priority, the lower the cost weight. The multi-level path evaluation standard includes three levels: safety evaluation, efficiency evaluation, and reliability evaluation. For each level, the corresponding evaluation indicators and thresholds are set to form an evaluation system.

[0027] The three-layer progressive path planner decomposes the evaluation criteria. At the global level, the improved A* algorithm is used to generate the initial path; at the local level, the dynamic spatiotemporal RRT* algorithm is used for path optimization; at the execution level, the executable motion trajectory is generated through trajectory smoothing. Through this layered processing, a set of path candidates is formed, each of which contains a sequence of path points and motion parameters. When evaluating the path candidate set, a collision risk assessment is performed to check whether the path has potential collisions with static obstacles or dynamic targets. The travel time calculation takes into account the vehicle's dynamic constraints and traffic rules restrictions to derive the expected travel time for each path. The optimal main path is selected through comprehensive evaluation.

[0028] According to the evaluation results of the main path, the search range of the alternative path is determined. A search area is established around the main path to generate multiple alternative paths, which form a path planning scheme with the main path. The generation of alternative paths fully considers the dynamic changes in road conditions and emergency needs in unexpected situations.

[0029] For example: autonomous vehicles need to transport containers from the terminal loading and unloading area to the yard storage area. The port area road network is divided into several nodes, including loading and unloading points, intersections, and yard entrances, and the connection relationship between nodes is established through topological structure analysis. The initial cost of all feasible paths is calculated based on the basic data of the road network, among which there are multiple optional paths from the loading and unloading area to the yard. Considering that it is currently the peak operation period, the main roads are given a higher priority for traffic, and the weight coefficient of the cost function is adjusted accordingly. The three-layer progressive planner generates a global path with the shortest distance, and then performs local optimization according to the real-time traffic conditions to generate a smooth motion trajectory. Through collision risk assessment, it is found that a section of the main path is undergoing road maintenance, so alternative paths are searched in the surrounding area to form a complete path planning solution with one main and two backups.

[0030] In a specific embodiment, the process of executing step S103 may specifically include the following steps: (1) Extract the spatiotemporal parameter sequence of the path from the path planning scheme and establish a dynamic target observation dataset in the port area; (2) Group and label the observed data set according to the target type to generate the target behavior feature vector; (3) Input the target behavior feature vector into the temporal attention network to extract the target movement law representation; (4) Construct a traffic flow density distribution map of the port area based on the target movement law representation and mark the high-risk intersection areas; (5) Conduct time-series correlation analysis on high-risk intersection areas and generate target conflict warning index; (6) Combine the target conflict warning index with the motion law representation to obtain real-time situation assessment data.

[0031] Specifically, the path spatiotemporal parameter sequence is extracted from the path planning scheme to establish a dynamic target observation data set for the port area. The path spatiotemporal parameter sequence refers to a serialized data structure consisting of the key point location information and the expected passing time on the vehicle's driving path, including latitude and longitude coordinates, timestamp, vehicle speed, direction angle and other parameters. By structuring these parameters to form a standardized data table, combined with the existing dynamic target information in the port area, such as the location and motion status of other vehicles, equipment and operators, a dynamic target observation data set for the port area is established.

[0032] After the observed data set is grouped and labeled according to the target type, it is necessary to generate a target behavior feature vector. The target types include forklifts, tractors, gantry cranes, bridge cranes, operators and other different categories, and each type of target has different motion characteristics. By analyzing the historical trajectory data of each type of target, characteristic parameters such as speed distribution, acceleration change, steering preference, and dwell time are extracted to form a multi-dimensional feature vector. For example, for a forklift type target, its feature vector may contain numerical representations of multiple dimensions such as average speed, maximum acceleration, common path, and operating area range.

[0033] Inputting the target behavior feature vector into the temporal attention network and extracting the target motion law representation is a key step. The temporal attention network is a deep learning network structure that specializes in processing time series data. It gives higher weights to key time points in the input sequence through the attention mechanism, thereby capturing the temporal dependencies of the target motion. The network contains an encoder-decoder structure. The encoder converts the input feature vector into a hidden layer representation. The attention layer calculates the weight distribution of different time steps, and the decoder generates a prediction result based on the attention weight. Through this network processing, the motion law representation of various targets can be extracted, including regular paths, typical speed curves, inertial motion patterns, etc.

[0034] Constructing a traffic flow density distribution map in the port area based on the target movement law representation and marking high-risk intersection areas are the basis for achieving safety warning. The traffic flow density distribution map is a data visualization structure in the form of a heat map. By gridding the port area, the target density of each grid unit in different time periods is calculated to form a dynamically changing density field. The density field is analyzed spatiotemporally to identify areas with frequent traffic flow intersections and high congestion possibilities, and risk thresholds are set based on historical accident data and expert experience to mark high-risk intersection areas.

[0035] Conducting time series correlation analysis on high-risk intersection areas and generating target conflict warning index is an important means of preventing accidents. Time series correlation analysis evaluates potential conflict risks by calculating the probability of different targets appearing simultaneously in the same area and the time overlap. The target conflict warning index is a comprehensive risk assessment indicator, which is calculated by weighted fusion of parameters such as target density, relative speed, predicted trajectory intersection angle, and minimum safety distance. The higher the index value, the greater the conflict risk and the stricter the preventive measures to be taken. The target conflict warning index is jointly analyzed with the motion law representation to obtain real-time situation assessment data. In the joint analysis process, the conflict warning index is used as the main risk indicator, combined with the uncertainty assessment in the target motion law representation and the historical abnormal behavior statistics, to construct a risk assessment model. The real-time situation assessment data contains multi-dimensional information such as risk level, risk duration, risk change trend, and recommended response measures, providing a basis for subsequent control decisions.

[0036] For example, in an autonomous driving system at a container port, the system extracts the spatiotemporal parameter sequence of an automatic transport vehicle (ATV) from the path planning scheme, including the coordinates and estimated passing time of 25 key path points. At the same time, the position and motion data of 10 forklifts, 8 tractors and 5 operators in the surrounding area are collected to form a dynamic target observation data set. After grouping and labeling the observation data set, multiple groups of feature vectors are generated. For example, the feature vector of the forklift group contains parameters such as average speed 3.5m / s, maximum acceleration 0.8m / s², and typical working area radius 32 meters. After these feature vectors are input into the temporal attention network, the motion law representations of various targets are extracted, such as the probability of a forklift to slow down and avoid at the edge of the container yard. Based on these motion law representations, the system constructs a traffic flow density distribution map of the port area and identifies three high-risk intersection areas where the main road and the branch line intersect. Through time series correlation analysis, the system calculated that the target conflict warning index of the No. 2 intersection area during the period of 10:30-11:00 in the morning reached 0.78 (0-1 interval), exceeding the high risk threshold of 0.75. The joint analysis results showed that the area needed to initiate a secondary safety response strategy, including reducing the speed of vehicles passing through the area and increasing the safety distance, forming real-time situation assessment data.

[0037] In a specific embodiment, the process of executing step S104 may specifically include the following steps: (1) Decompose the real-time situation assessment data, extract vehicle state parameters and motion constraints, and generate a vehicle control parameter set; (2) Grouping vehicle control parameter sets according to vehicle task priorities and constructing a vehicle collaborative decision space; (3) Match the control variables in the vehicle collaborative decision space with constraints to generate a vehicle collaborative control constraint matrix; (4) Optimizing the vehicle cooperative control constraint matrix through the distributed model predictive control algorithm to form a vehicle action sequence table; (5) Conflict detection and avoidance rule verification based on the vehicle action sequence table, and establishment of a vehicle command priority list; (6) Sort and merge the vehicle command priority list in chronological order to generate a vehicle control command sequence.

[0038] Specifically, the real-time situation assessment data is decomposed, the vehicle state parameters and motion constraints are extracted, and the vehicle control parameter set is generated. The real-time situation assessment data is a multidimensional information structure, which contains the vehicle's current position, speed, acceleration and other state parameters, as well as the safety distance of the surrounding environment, obstacle distribution and other constraint information. The data decomposition process uses a feature separation algorithm to divide the original situation data into three categories according to physical properties: dynamic parameter group, environmental constraint group and task requirement group. The dynamic parameter group includes speed limit, steering angle limit, acceleration boundary, etc.; the environmental constraint group includes obstacle avoidance distance, safety boundary, traffic rules, etc.; the task requirement group includes target position, arrival time, energy consumption requirements, etc. Through data reorganization and normalization processing, these parameters are integrated into a standardized vehicle control parameter set. Grouping the vehicle control parameter set according to the vehicle task priority and constructing a vehicle collaborative decision space is the basis for multi-vehicle collaboration. The vehicle task priority is a priority level comprehensively evaluated based on factors such as task urgency, importance, and resource occupancy. It is usually divided into four levels: emergency tasks, important tasks, routine tasks, and background tasks. The grouping process uses a hierarchical clustering method to aggregate vehicle control parameter sets with similar priorities at the same decision level to form a multi-level decision structure. The vehicle collaborative decision space is a multi-dimensional vector space representation. Each dimension represents a decision variable, such as speed selection, path selection, action timing, etc. All possible decision combinations constitute the decision space. By mapping decision variables of different priorities to corresponding spatial regions, a hierarchical collaborative decision space is established, which provides a basis for subsequent constraint matching.

[0039] Matching the control variables in the vehicle collaborative decision space with constraints and generating the vehicle collaborative control constraint matrix involves a complex constraint processing process. Control variables include directly operable variables such as speed, acceleration, steering angle, and braking force, and constraints include physical limit constraints, safety constraints, energy consumption constraints, and other types. Constraint matching uses a graph-based mapping algorithm to establish a connection relationship between each control variable and the relevant constraints to form a constraint network. By analyzing the constraint network, key constraints and redundant constraints are identified, and the constraint structure is optimized. The vehicle collaborative control constraint matrix is ​​a mathematical representation. Each element in the matrix represents the constraint coefficient of a specific control variable under a specific constraint condition, forming a set of constraint equations.

[0040] The core control step is to optimize the vehicle cooperative control constraint matrix through the distributed model predictive control algorithm to form a vehicle action sequence table. The distributed model predictive control algorithm is an advanced control algorithm that predicts the behavior of the system in the future and optimizes the current control decision based on the prediction results. First, a vehicle dynamics model is established to predict the future state trajectory under different control inputs. Then the objective function is set, including multiple optimization goals such as trajectory tracking error, control input smoothness, and energy consumption minimization. By solving the constrained optimization problem, the optimal control sequence that meets all constraints is obtained. The distributed processing architecture enables multiple vehicle controllers to calculate in parallel and exchange intermediate results through communication to achieve global optimization. The optimized calculation results form a vehicle action sequence table, which includes specific speed curves, steering angle changes, acceleration and deceleration timing, and other time-series action instructions.

[0041] Based on the vehicle action sequence table, conflict detection and avoidance rule verification, and establishing a vehicle instruction priority list are key links to ensure safety. Conflict detection uses a spatiotemporal trajectory cross analysis method to identify potential conflict points by calculating the minimum distance and time overlap of different vehicle predicted trajectories. For detected conflicts, preset avoidance rules are applied to handle them, such as right-hand vehicle priority, straight-moving vehicle priority, and loaded vehicle priority. During the avoidance rule verification process, the corresponding rules are applied to each potential conflict to determine the yielding vehicle and priority vehicle, and correct the corresponding action sequence. The vehicle instruction priority list records the execution priority and mutual dependence of all instructions to ensure that the control instructions can be executed in a reasonable order in the event of a conflict. The time sequence sorting uses a topological sorting algorithm to determine the final execution order of the instructions based on the dependency relationship and execution time point between the instructions. During the instruction merging process, instructions with similar time and similar operations are merged and simplified to reduce the switching frequency of control instructions and improve the stability of control. The vehicle control instruction sequence is a standardized instruction set sent directly to the vehicle actuator, which contains specific execution time, execution content and execution parameters.

[0042] For example, in an autonomous driving system at a container port, when multiple autonomous transport vehicles need to work together between container yards, the system first obtains real-time situation assessment data, including the current status and surrounding environment information of five vehicles. After decomposing these data, the state parameters such as the position, speed, acceleration, and other state parameters of each vehicle, as well as constraints such as road width, turning radius, and safety distance are extracted to form a vehicle control parameter set. Subsequently, the five vehicles are divided into two high-priority vehicles (carrying important goods) and three regular-priority vehicles according to the urgency of the task, and a collaborative decision space is constructed. In the decision space, the control variables such as speed, path, and scheduling time are matched with constraints to generate a collaborative control constraint matrix. The optimal control strategy is calculated through a distributed model predictive control algorithm to form a sequence table containing detailed actions within the next 30 seconds. The system then detects that there is a potential intersection conflict between two vehicles in a narrow section of road, applies the "heavy-loaded vehicle priority" avoidance rule, adjusts the speed curve of the regular-priority vehicle, and delays its passing time, thereby establishing an instruction priority list. Ultimately, the system sorted and merged these instructions in chronological order to generate a highly optimized vehicle control instruction sequence, enabling all vehicles to complete collaborative tasks efficiently and safely, greatly improving port operation efficiency.

[0043] In a specific embodiment, the process of executing step S105 may specifically include the following steps: (1) Construct an instruction execution timing table based on the vehicle control instruction sequence, extract vehicle motion control parameters, and generate vehicle motion state monitoring conditions; (2) Input the vehicle motion state monitoring conditions into the adaptive safety constraint module to establish a dynamic safety boundary constraint set; (3) Perform boundary detection on the vehicle’s current motion state based on the dynamic safety boundary constraint set and generate safety risk assessment indicators; (4) Using security risk assessment indicators as weight coefficients, a hierarchical security response rule base is constructed to form a multi-level security response strategy; (5) Pair the multi-level safety response strategy with historical control data, input it into the deep reinforcement learning network, and generate control strategy optimization parameters; (6) Dynamically adjust the multi-level safety response strategy through control strategy optimization parameters to complete the vehicle's motion status monitoring.

[0044] Specifically, the timing table contains the execution time, execution order and execution conditions of each instruction, and extracts motion control parameters such as speed instructions, steering instructions, and braking instructions. By analyzing the time distribution and spatial distribution characteristics of these parameters, the motion state monitoring conditions are generated, including specific constraint values ​​such as speed threshold, steering angle range, and acceleration limit. After the motion state monitoring conditions are input into the adaptive safety constraint module, a dynamic safety boundary constraint set is established in combination with the vehicle dynamics characteristics and road conditions. The constraint set consists of speed safety boundary, position safety boundary and posture safety boundary. The speed safety boundary defines the safe speed range of the vehicle in different scenarios, the position safety boundary determines the minimum safe distance from other vehicles and fixed facilities, and the posture safety boundary limits the yaw angle and roll angle of the vehicle.

[0045] When performing boundary detection on the current motion state of the vehicle, the actual motion parameters are compared with the safety boundary constraint set in real time. The deviation value between each monitoring parameter and the corresponding boundary is calculated, the severity of each deviation is comprehensively evaluated, and a quantitative safety risk assessment index is generated. This index reflects the safety level of the current motion state of the vehicle. The larger the value, the higher the risk. Based on the safety risk assessment index, a safety response rule base containing multiple response levels is constructed. Different risk indicator intervals correspond to different response strategies, forming a multi-level safety response strategy. The response levels from low to high are normal operation, early warning prompts, active intervention and emergency braking. Each level has clear trigger conditions and response measures.

[0046] The multi-level safety response strategy is paired with historical control data for analysis, and evaluation indicators of the strategy execution effect are extracted, including response timeliness, control stability, and safety margin. These data are input into the deep reinforcement learning network as training samples. Through repeated learning and optimization, the network generates control strategy optimization parameters that can adapt to different scenarios. The control strategy optimization parameters are used to dynamically adjust the multi-level safety response strategy. The adjustment content includes the trigger threshold, response intensity, and execution timing of each level, ensuring that the safety response strategy can be flexibly adjusted according to the actual situation and realize accurate monitoring of the vehicle's motion status.

[0047] For example, when an autonomous container truck is running in a narrow passage, a timing table is established according to the control instruction sequence to extract key parameters such as steering angle and driving speed. Taking into account the width of the passage and the road conditions, the adaptive safety constraint module sets a smaller speed boundary and a larger safety distance requirement. During driving, real-time boundary detection found that the distance between the vehicle and the container pile on one side was close to the safety boundary, and the safety risk assessment index increased. The multi-level safety response strategy immediately activated the warning level, and the deep reinforcement learning network calculated the optimized steering angle and deceleration strategy based on historical data. The dynamically adjusted response strategy enables the vehicle to smoothly adjust the driving route and always stay within the safety boundary. The data processing and decision-making adjustments in the entire monitoring process are closely centered on the core goal of safety, and each link maintains a strict logical connection.

[0048] In a specific embodiment, the process of executing step S106 may specifically include the following steps: (1) Classify and sort the safety risk assessment indicators according to risk levels and generate a risk classification threshold table; (2) Divide the values ​​in the risk classification threshold table into intervals and establish a response trigger condition matrix; (3) Construct a security response rule mapping set based on the response trigger condition matrix and mark the rule priority number; (4) Associating the rule priority number with the vehicle control constraint condition to form a rule constraint relationship table; (5) According to the rule constraint relationship table, the response rules are hierarchically combined to build a security response rule library; (6) Prioritize the rule priority numbers according to the security response rule base to form a multi-level security response strategy.

[0049] Specifically, starting from the classification and sorting of safety risk assessment indicators, the risk assessment indicators are divided into different levels according to the degree of danger. The risk assessment indicators include multi-dimensional indicators such as vehicle motion state parameters (such as speed, acceleration, steering angle), environmental perception data (such as obstacle distance, relative speed) and task execution status (such as loading and unloading status, battery power).

[0050] For speed indicators, the benchmark speed threshold is set at 20km / h in the port scenario. When the vehicle speed exceeds this threshold, it enters the risk warning state. The benchmark threshold of the acceleration indicator is set to 2m / s², and the threshold of the steering angle change rate is 30° / s. These kinematic parameters are combined with environmental perception data such as obstacle distance and relative speed to form a multi-dimensional risk assessment system. For example, when an autonomous driving vehicle detects a dynamic obstacle within 10 meters ahead and the relative speed exceeds 5km / h, a higher risk level will be triggered. Based on these specific indicators, a threshold table containing multiple risk levels is generated. The risk classification threshold table is then divided into intervals to establish a response trigger condition matrix. The interval division adopts a non-uniform division method based on the degree of risk, and differentiated trigger thresholds are set for different risk indicators. Taking vehicle speed as an example, 20-25km / h is defined as the first-level warning interval, 25-30km / h is the second-level warning interval, and more than 30km / h is the third-level warning interval. The obstacle distance is also divided into three intervals: 10-15m, 5-10m, and 0-5m. These interval division results are multiplied by the weight coefficients of each indicator to form a response trigger condition matrix.

[0051] Based on the response trigger condition matrix, a safety response rule mapping set is constructed and the rule priority number is marked. The mapping rules are set based on the actual scenario requirements. For example, when a dynamic obstacle is detected within 5 meters ahead, the emergency braking rule is triggered and the rule is assigned the highest priority number 1; when the vehicle speed is detected to exceed 25km / h, the deceleration rule is triggered and assigned the priority number 2. In this way, a clear mapping relationship is established between various risk scenarios and corresponding control responses. The rule priority number is associated with the specific vehicle control constraint conditions to form a rule constraint relationship table. The control constraint conditions include parameters such as maximum braking deceleration, maximum steering angular velocity, and minimum safe distance. For example, the maximum braking deceleration corresponding to the emergency braking rule (priority 1) is set to 3m / s², and the minimum safe distance is set to 2 meters; the braking deceleration corresponding to the deceleration rule (priority 2) is set to 1m / s². Through this association, the abstract rule number is converted into specific control parameters. According to the rule constraint relationship table, the response rules are hierarchically combined to construct a safety response rule base. The rule combination takes into account the coupling effect of multiple risk factors. For example, when there are two risk factors of high speed and small distance at the same time, a more stringent control parameter is selected. The rule base contains three levels: basic security rules, warning rules and emergency rules, and multiple specific response rules are set under each level.

[0052] According to the security response rule base, the rule priority numbers are sorted to form a multi-level security response strategy. Priority sorting adopts a comprehensive evaluation method based on risk level and timeliness to ensure that the most suitable response strategy can be triggered in a timely manner in complex scenarios. For example, when an autonomous vehicle encounters a sudden obstacle on the port channel, the response strategy will activate multiple levels of control instructions such as deceleration, avoidance, and braking in sequence. Each control instruction has a clear trigger condition and execution priority.

[0053] For example, when an autonomous container transporter is driving on the main road of the port area, a forklift suddenly appears in front of it, which is performing loading and unloading operations. At this time, the safety monitoring module detects that the distance to the forklift is 8 meters, the relative speed is 10km / h, and the vehicle speed is 22km / h. Based on the risk classification threshold table, this scenario triggers two risk indicators, distance (5-10m interval) and speed (20-25km / h interval). The response trigger condition matrix combines the two risk factors to determine the need to perform deceleration and avoidance operations. The safety response rule mapping set maps the scenario to the avoidance rule (priority 3) and the deceleration rule (priority 2). The rule constraint relationship table assigns specific control parameters to these two rules: deceleration 1.5m / s², steering angular velocity not exceeding 20° / s. The multi-level safety response strategy performs deceleration and avoidance actions in order of priority to ensure that the transporter can safely pass through the section. The entire process reflects the complete logical chain from risk identification to the generation of specific control instructions, and fully considers the safe operation requirements of autonomous vehicles in the port area scenario.

[0054] The above describes the path planning and control method of the container port autonomous driving vehicle in the embodiment of the present application. The following describes the path planning and control system of the container port autonomous driving vehicle in the embodiment of the present application. Figure 2 In the embodiments of the present application, one embodiment of the path planning and control system for the container port autonomous driving vehicle includes: The fusion module is used to collect port area environmental data through a multi-source sensor network, and fuse the environmental data through a spatiotemporal registration algorithm to obtain a digital twin environment model of the port area; The hierarchical module is used to perform hierarchical optimization processing on the vehicle path through a three-layer progressive path planner according to the port area digital twin environment model, and obtain a path planning solution including a main path and alternative paths; The prediction module is used to perform trajectory prediction analysis on dynamic targets in the port area based on the path planning scheme using the temporal attention network to obtain real-time situation assessment data; A control module, configured to perform coordinated control processing on the vehicle through a distributed model predictive control algorithm based on the real-time situation assessment data to generate a vehicle control instruction sequence; The monitoring module is used to monitor the vehicle's motion state in real time through the adaptive safety constraint module according to the vehicle control instruction sequence, form a multi-level safety response strategy, and continuously optimize the control strategy using a deep reinforcement learning network.

[0055] Through the collaboration of the above components, the port area environmental data is collected through a multi-source sensor network, and the spatiotemporal registration algorithm is used for fusion processing to build a high-precision digital twin environment model, which effectively solves the problem of incomplete and inaccurate data from traditional single sensors and significantly improves the comprehensiveness and accuracy of environmental perception. The three-layer progressive path planner is used to optimize the vehicle path in layers, which can not only generate the optimal main path, but also include multiple alternative paths, enhance the system's ability to respond to emergencies and improve the robustness of path planning. The introduction of the temporal attention network enables the system to accurately predict the trajectory of dynamic targets in the port area. Through deep learning and temporal correlation analysis of target motion characteristics, the system's prediction accuracy of dynamic environmental changes is greatly improved. The application of the distributed model predictive control algorithm realizes the collaborative control of multiple vehicles, and can dynamically adjust the control strategy based on real-time situation assessment data, significantly improving the efficiency and safety of group control. The adaptive safety constraint module cooperates with the continuous optimization mechanism of the deep reinforcement learning network, so that the system can continuously improve the control strategy according to the actual operation data, and establish an adaptive evolutionary safety control system. Especially in terms of algorithm features, the temporal attention network extracts the key temporal features of dynamic targets through the attention mechanism processing of time series data, making trajectory prediction more accurate; the deep reinforcement learning network continuously improves the intelligence and adaptability of control decisions through continuous interaction with the environment and strategy optimization. The comprehensive application of these algorithm features significantly enhances the system's adaptability and control performance in complex port environments. Overall, the solution has built an intelligent and adaptive port autonomous driving control system through the organic combination of multiple innovative algorithms, achieving all-round optimization of environmental perception, path planning, trajectory prediction, collaborative control and safety monitoring, and greatly improving the operating efficiency and safety of port autonomous driving vehicles.

[0056] Reference Figure 3 In an embodiment of the present invention, a computer device is also provided. The computer device may be a server, and its internal structure may be as follows: Figure 3As shown. The computer device includes a processor, a memory, a display screen, an input device, a network interface and a database connected through a system bus. Among them, the processor designed by the computer is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store the corresponding data in this embodiment. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, the above method is implemented.

[0057] Those skilled in the art will understand that Figure 3 The structure shown in the figure is merely a block diagram of a portion of the structure related to the solution of the present invention, and does not constitute a limitation on the computer device to which the solution of the present invention is applied.

[0058] An embodiment of the present invention further provides a computer-readable storage medium on which a computer program is stored, and when the computer program is executed by a processor, the above method is implemented. It can be understood that the computer-readable storage medium in this embodiment can be a volatile readable storage medium or a non-volatile readable storage medium.

[0059] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media provided by the present invention and used in the embodiments may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double-speed data rate SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM.

[0060] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described systems, systems and units can refer to the corresponding processes in the aforementioned method embodiments 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 application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. 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 application.

Claims

1. A path planning and control method for an autonomous driving vehicle in a container port, characterized in that: The path planning and control method of the container port autonomous driving vehicle includes: The port area environmental data is collected through a multi-source sensor network, and the environmental data is fused and processed through a spatiotemporal registration algorithm to obtain a digital twin environmental model of the port area; According to the digital twin environment model of the port area, the vehicle path is hierarchically optimized through a three-layer progressive path planner to obtain a path planning solution including a main path and alternative paths; Based on the path planning scheme, the temporal attention network is used to perform trajectory prediction analysis on dynamic targets in the port area to obtain real-time situation assessment data; Based on the real-time situation assessment data, the vehicle is collaboratively controlled by a distributed model predictive control algorithm to generate a vehicle control instruction sequence; According to the vehicle control instruction sequence, the vehicle's motion state is monitored in real time via an adaptive safety constraint module to form a multi-level safety response strategy, and the control strategy is continuously optimized using a deep reinforcement learning network.

2. The path planning and control method for an autonomous container port vehicle according to claim 1, characterized in that: The port area environmental data is collected through a multi-source sensor network, and the environmental data is fused and processed through a spatiotemporal registration algorithm to obtain a port area digital twin environment model, including: In the step of collecting port area environmental data via the multi-source sensor network, the environmental data is divided into a static environmental data stream and a dynamic environmental data stream; Performing spatial coordinate analysis on the static environment data stream to establish a port infrastructure location index table; Performing target identification and classification on the dynamic environment data stream to generate real-time target status records; Pairing the port infrastructure location index table with the real-time target status record according to timestamps to form a time series environment feature matrix; Performing data cleaning and outlier processing on the time series environmental feature matrix to construct a standardized environmental data set; The standardized environmental data set is input into the spatiotemporal registration algorithm, and the digital twin environmental model of the port area is generated through feature mapping and data alignment.

3. The path planning and control method for an autonomous container port vehicle according to claim 1, characterized in that: According to the port area digital twin environment model, the vehicle path is hierarchically optimized through a three-layer progressive path planner to obtain a path planning solution including a main path and an alternative path, including: Parsing the port area digital twin environment model into a road network topology diagram, marking the vehicle traffic attributes of the road network nodes, and generating a road network basic data set; Extracting the node distance matrix and the turning constraint conditions from the road network basic data set to construct an initial cost function for path planning; The initial cost function of the path planning is weighted according to the vehicle traffic priority to generate a multi-level path evaluation standard; Decomposing the multi-level path evaluation criteria via the three-layer progressive path planner to form a path candidate set; Performing collision risk assessment and travel time calculation on the candidate path set to select the main path; The alternative path search range is determined according to the evaluation result of the main path, and the path planning solution is generated.

4. The path planning and control method for an autonomous container port vehicle according to claim 1, characterized in that: Based on the path planning scheme, the temporal attention network is used to perform trajectory prediction analysis on the dynamic targets in the port area to obtain real-time situation assessment data, including: Extracting a path spatiotemporal parameter sequence from the path planning scheme and establishing a dynamic target observation data set for the port area; The observation data set is grouped and labeled according to target type to generate a target behavior feature vector; Inputting the target behavior feature vector into the temporal attention network to extract the target movement law representation; Construct a traffic flow density distribution map of the port area according to the target movement law representation, and mark the high-risk intersection area; Conducting time series correlation analysis on the high-risk intersection area to generate a target conflict warning index; The target conflict warning index and the motion law representation are jointly analyzed to obtain the real-time situation assessment data.

5. The path planning and control method for an autonomous container port vehicle according to claim 1, characterized in that: The method of performing coordinated control processing on the vehicle based on the real-time situation assessment data through a distributed model predictive control algorithm to generate a vehicle control instruction sequence includes: Decomposing the real-time situation assessment data, extracting vehicle state parameters and motion constraints, and generating a vehicle control parameter set; The vehicle control parameter sets are grouped according to vehicle task priorities to construct a vehicle collaborative decision space; Performing constraint matching on the control variables in the vehicle collaborative decision space to generate a vehicle collaborative control constraint matrix; Optimizing and calculating the vehicle cooperative control constraint matrix via the distributed model predictive control algorithm to form a vehicle action sequence table; Perform conflict detection and avoidance rule verification based on the vehicle action sequence table, and establish a vehicle instruction priority list; The vehicle command priority list is sorted and merged in chronological order to generate the vehicle control command sequence.

6. The path planning and control method for an autonomous container port vehicle according to claim 1, characterized in that: According to the vehicle control instruction sequence, the vehicle's motion state is monitored in real time via an adaptive safety constraint module to form a multi-level safety response strategy, and the control strategy is continuously optimized using a deep reinforcement learning network, including: Building an instruction execution timing table based on the vehicle control instruction sequence, extracting vehicle motion control parameters, and generating vehicle motion state monitoring conditions; Inputting the vehicle motion state monitoring condition into the adaptive safety constraint module to establish a dynamic safety boundary constraint set; Perform boundary detection on the current motion state of the vehicle according to the dynamic safety boundary constraint set to generate a safety risk assessment index; Using the security risk assessment index as a weight coefficient, constructing a hierarchical security response rule base to form the multi-level security response strategy; Pairing the multi-level safety response strategy with historical control data, inputting the data into the deep reinforcement learning network, and generating control strategy optimization parameters; The multi-level safety response strategy is dynamically adjusted via the control strategy optimization parameters to complete the motion state monitoring of the vehicle.

7. The path planning and control method for an autonomous container port vehicle according to claim 6, characterized in that: The method of using the security risk assessment index as a weight coefficient to construct a hierarchical security response rule base and form the multi-level security response strategy includes: Classify and sort the security risk assessment indicators according to risk levels to generate a risk classification threshold table; Divide the values ​​in the risk grading threshold table into intervals to establish a response trigger condition matrix; Construct a security response rule mapping set based on the response trigger condition matrix and mark the rule priority sequence number; Associating the rule priority sequence number with the vehicle control constraint condition to form a rule constraint relationship table; According to the rule constraint relationship table, the response rules are hierarchically combined to build a security response rule library; The rule priority numbers are prioritized according to the security response rule base to form the multi-level security response strategy.

8. A path planning and control system for an autonomous container port vehicle, used to implement the path planning and control method for an autonomous container port vehicle as claimed in any one of claims 1 to 7, characterized in that: The path planning and control system of the container port autonomous driving vehicle includes: A fusion module is used to collect port area environmental data through a multi-source sensor network, and to fuse the environmental data through a spatiotemporal registration algorithm to obtain a digital twin environmental model of the port area; A hierarchical module, used to perform hierarchical optimization processing on the vehicle path through a three-layer progressive path planner according to the port area digital twin environment model, and obtain a path planning solution including a main path and an alternative path; A prediction module, for performing trajectory prediction analysis on dynamic targets in the port area based on the path planning scheme using a temporal attention network to obtain real-time situation assessment data; A control module, configured to perform coordinated control processing on the vehicle through a distributed model predictive control algorithm based on the real-time situation assessment data to generate a vehicle control instruction sequence; The monitoring module is used to monitor the vehicle's motion state in real time through the adaptive safety constraint module according to the vehicle control instruction sequence, form a multi-level safety response strategy, and continuously optimize the control strategy using a deep reinforcement learning network.

9. A computer device, characterized in that: It comprises a memory and a processor, wherein the memory stores a computer program that can be run on the processor, and is characterized in that when the processor executes the computer program, the path planning and control method for the container port autonomous driving vehicle described in any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the processor executes the path planning and control method for an autonomous container port vehicle according to any one of claims 1 to 7.

Images