Multi-tractor collaborative operation scheduling system for closed park and control method

Abstract

The invention belongs to the technical field of agricultural automation, and provides a closed park-oriented tractor multi-machine collaborative operation scheduling system and a control method, and the method comprises the steps: employing a multi-source heterogeneous sensor network to construct an environment perception layer, generating an environment state tensor and a dynamic environment model through space-time registration and data fusion, and obtaining a dynamic environment model; a complex task is decoupled into an atomic operation unit, a single tractor optimal task sequence is generated by utilizing dynamic planning, a genetic algorithm is improved to realize global operation distribution, and an optimal path is planned for tractors in a closed park through an ant colony algorithm in combination with a dynamic environment model to control the tractors to perform collaborative operation. Deploying an edge computing node to receive the tractor pose data in real time, judging an actual conflict risk by calculating space-time conflict density, fitting a prediction trajectory function to detect a prediction conflict risk, and if the actual conflict risk or the prediction conflict risk exists, triggering conflict coordination; and selecting a hierarchical coordination strategy for the tractor according to the conflict risk type to realize obstacle avoidance control.

Description

Technical Field

[0001] The present invention belongs to the technical field of agricultural automation, and more specifically, it is a multi-tractor collaborative operation scheduling system and control method for closed campuses. Background Art

[0002] With the acceleration of the agricultural modernization process, the scale of agricultural production in closed campuses has been continuously expanding, and the operation tasks have become increasingly complex. The traditional manual operation or single-machine operation mode has been difficult to meet the requirements of efficient and precise agricultural production. In a closed campus, there are usually multiple operation links such as land tillage, sowing, fertilization, irrigation, and harvesting. These links are interrelated and have strict timing requirements. Therefore, realizing the multi-tractor collaborative operation in a closed campus and improving the operation efficiency and quality have become an urgent need for the development of agricultural modernization.

[0003] In the prior art, the environmental perception mainly relies on a single sensor, making it difficult to comprehensively obtain multi-dimensional information such as terrain slope, soil compaction, and dynamic obstacles, resulting in a static and one-sided environmental model. When dynamic obstacles enter the operation area, the existing system lacks a real-time prediction and update mechanism, which is prone to cause path planning failure or safety accidents. On the other hand, in the prior art, the task scheduling lacks the processing of multi-dimensional constraints. Complex agricultural operations include multiple links such as tillage, sowing, and fertilization. Each task has strict timing dependencies, resource constraints, and equipment load balancing requirements. The existing methods often simply allocate tasks without constructing a task dependency relationship model, resulting in chaotic operation sequences, frequent implement replacements, or uneven tractor loads, significantly reducing the operation efficiency. When multiple tractors operate collaboratively, the prior art relies on simple distance threshold detection for immediate collision risks, lacking the ability to predict future trajectory intersections, unable to give early warnings and plan obstacle avoidance paths, and only braking emergently when the distance is less than the safety threshold, resulting in operation interruptions or efficiency losses.

[0004] In view of the above problems, the present invention proposes a multi-tractor collaborative operation scheduling system and control method for closed campuses. Summary of the Invention

[0005] In order to make up for the deficiencies of the prior art and solve at least one of the technical problems in the background art.

[0006] The technical solution adopted by the present invention to solve the technical problems is as follows: A multi-tractor collaborative operation scheduling system for closed campuses, comprising: An environmental modeling module: generating an environmental state tensor using a multi-source heterogeneous sensor network to construct a dynamic environmental model; Path planning module: decouple complex operation tasks into atomic operation units, use dynamic programming to generate the optimal task sequence for a single tractor, combine with an improved genetic algorithm to achieve global task allocation for multiple tractors, and integrate the ant colony algorithm with the dynamic environment model to plan the optimal path for each tractor; Decouple complex operation tasks in a closed park into several smallest executable atomic operation units, describe the characteristics of each atomic operation unit through multi-dimensional feature vectors, and construct a directed acyclic graph of operation dependencies between atomic operation units. Combine with the topological sorting algorithm to generate the initial operation sequence; Combine the initial operation sequence with the environmental state tensor, use the dynamic programming algorithm, define the state space to represent the minimum cumulative cost when processing each atomic operation unit, and iteratively solve the minimum cumulative cost through the Bellman equation to generate the optimal task sequence for a single tractor; Obtain the optimal operation sequence of a single tractor, and achieve global operation allocation for tractors in a closed park through an improved genetic algorithm. In the improved genetic algorithm, design a multi-objective fitness function to balance efficiency, balance, and safety, and adopt a two-layer coding structure. The upper layer is the allocation layer, which generates the optimal allocation matrix that specifies the tractor corresponding to each task, and the lower layer is the sequence layer, which generates the atomic operation unit sequence of each tractor; Obtain the atomic operation unit sequence of the tractor, extract the geometric center coordinates of the operation area of each atomic operation unit to form a node set, integrate the environmental characteristics and geometric distance to calculate the edge weights between nodes, and combine the nodes and edge weights to obtain a task node graph with environmental weights; Perform pheromone initialization and ant colony behavior modeling in the ant colony algorithm, set the core parameters, define the state transition rule to calculate the state transition probability of each ant choosing the next node at any node, output the pheromone matrix, and set each ant to traverse all nodes according to the state transition probability to generate a complete path; Obtain and calculate the total length of the complete path according to the task node graph with environmental weights, iteratively update the pheromone matrix until the optimal path is found and the iteration is terminated, and output the obtained optimal path; Collision monitoring module: control the tractors to perform collaborative operations based on the optimal path, analyze the spatio-temporal collision density through a sliding time window according to the pose data collected by the edge computing node in real time, combine with the trajectory prediction function to detect the actual collision risk and predicted collision risk, and judge whether to trigger collision coordination; Establish a kinematic model to describe the motion state of the tractor, and control the tractors in the closed park to perform collaborative operations through the optimal path output by the path planning module; Five edge computing nodes are deployed at the four corners and the center of the closed park. They are equipped with industrial-grade hosts and support wireless communication to ensure that there are no blind spots in the entire area. The position data of each tractor is received in real time. The position data is obtained by integrating GNSS positioning with inertial navigation results, and the real-time updated environmental state tensor is obtained synchronously. Performing linear interpolation processing on the posture data of each tractor to generate a continuous posture sequence of each tractor; Construct a dynamic three-dimensional space-time relationship matrix, define a sliding time window, obtain the continuous posture sequence of each tractor, and calculate the Euclidean distance between any two tractors at each time step in the sliding time window. If the Euclidean distance is less than the safety distance, define the corresponding element in the dynamic three-dimensional space-time relationship matrix as 1, otherwise it is defined as 0; The spatiotemporal conflict density is calculated based on the dynamic three-dimensional spatiotemporal relationship matrix. If the spatiotemporal conflict density is greater than or equal to the conflict threshold, it is determined that there is an actual conflict risk between the two tractors. The continuous posture sequence of each tractor in the sliding time window is obtained for least squares fitting, and the predicted trajectory function of each tractor in the prediction window with the change time as the variable is obtained. The intersection equation of the predicted trajectory function between any two tractors is solved. If the equation has a positive real root, the positive real root represents the change time. If the change time is less than the time threshold, it is determined that there is a prediction conflict risk between the two tractors. Conflict coordination module: If conflict coordination is triggered, a hierarchical coordination strategy is selected for the tractor to achieve obstacle avoidance control; If an actual conflict risk is detected, the real-time speed adjustment strategy is triggered first, followed by the local path replanning strategy. If a predicted conflict risk is detected, the local path replanning strategy is triggered; In the real-time speed adjustment strategy, speed attenuation control is implemented for two tractors with actual conflict risks. If the movement speeds of both tractors at the current time point are less than the speed threshold, the local path replanning strategy is triggered. If the local path replanning strategy is triggered, based on the optimal path of the two tractors, the posture data of the two tractors at the end of the prediction window is calculated as the end point, and the posture data of the two tractors at the current time point is used as the starting point. The local optimal path is generated for the two tractors through the ant colony algorithm, and the path segment in the prediction window in the optimal path of the two tractors is replaced. The optimal path of the two tractors is updated, and the updated optimal path is used to control the tractors to work collaboratively.

[0007] The method for dispatching and controlling tractor multi-machine cooperative operation in a closed park includes the following steps: Step 1: Use a multi-source heterogeneous sensor network to generate an environmental state tensor and build a dynamic environmental model; Step 2: Decouple complex operation tasks into atomic operation units, generate the optimal task sequence for a single tractor using dynamic programming, implement global task allocation for multiple tractors in combination with an improved genetic algorithm, and fuse the ant colony algorithm with the dynamic environment model to plan the optimal path for each tractor; Step 3: Control the tractors to perform collaborative operations based on the optimal path. According to the pose data collected in real time by the edge computing nodes, analyze the spatio-temporal conflict density through a sliding time window, and combine the trajectory prediction function to detect the actual conflict risk and the predicted conflict risk, and determine whether to trigger conflict coordination; Step 4: If conflict coordination is triggered, select a hierarchical coordination strategy for the tractors to achieve obstacle avoidance control.

[0008] The beneficial effects of the present invention are as follows: 1. The present invention uses a multi-source heterogeneous sensor network to construct an environmental perception layer, generates a high-precision dynamic environment model through spatio-temporal registration and data fusion, enables the tractor to accurately perceive the complex environment of the closed park, decouples tasks and generates a job-dependent directed acyclic graph, plans the task sequence and allocation in combination with dynamic programming and an improved genetic algorithm, greatly optimizes the operation process, plans the path using the ant colony algorithm based on the dynamic environment model, enables the tractors to perform collaborative operations according to the optimal path, reduces repetitive operations and ineffective movements, significantly improves the operation accuracy and overall efficiency, and provides strong support for efficient production in the park.

[0009] 2. The present invention deploys edge computing nodes to receive pose data in real time, can timely determine the actual and predicted conflict risks through spatio-temporal conflict density calculation and predicted trajectory function. After triggering conflict coordination, the hierarchical coordination strategy accurately implements obstacle avoidance control according to the conflict type, effectively avoids collision accidents, and ensures the safety of tractors and personnel. At the same time, the dynamic environment model enables the system to adapt to changes in the park environment in real time, quickly adjusts the operation plan and path, enhances the flexibility and adaptability of multi-tractor collaborative operations, and ensures the stable and orderly development of operations in the closed park. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] The present invention will be further described below with reference to the accompanying drawings.

[0011] Figure 1 is the system module architecture diagram of the tractor multi-tractor collaborative operation scheduling system for a closed park according to an embodiment of the present invention; Figure 2 is the step flow chart of the tractor multi-tractor collaborative operation scheduling control method for a closed park according to an embodiment of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0012] In order to make the technical means, creative features, achieved purposes and effects of the present invention easy to understand, the present invention will be further described below in conjunction with specific embodiments.

[0013] Example 1, please refer to Figure 1 As shown in the figure, the multi-tractor collaborative operation scheduling system for a closed park according to the embodiment of the present invention includes the following modules: Environmental modeling module: A multi-source heterogeneous sensor network is used to construct an environmental perception layer. Through spatio-temporal registration and data fusion, a three-dimensional voxel grid is constructed in the global coordinate system to generate an environmental state tensor, and sensor data is synchronized in real time to generate a high-precision dynamic environmental model; A multi-source heterogeneous sensor network is used to construct an environmental perception layer. The multi-source heterogeneous sensor network includes lidar, visual cameras, soil sensors, and a GNSS positioning system. A high-precision environmental model is generated through spatio-temporal registration and data fusion technology; Specifically, the upper left corner of the closed park is selected as the origin, and a Universal Transverse Mercator coordinate system is established , with the Z-axis perpendicular upward, in meters. The rigid body transformation matrix of the lidar, visual cameras, and soil sensors relative to the GNSS positioning antenna is determined through a precision calibration board. The GNSS positioning data is converted into the absolute position in the global coordinate system through real-time kinematic technology. The multi-source heterogeneous sensor network is time-synchronized to ensure the alignment of multi-source data in the time dimension, and the data acquisition period is defined , and the multi-source heterogeneous sensor network synchronously acquires data triggered by the synchronized clock; The point cloud data collected by the lidar is subjected to outlier filtering to remove noise points. The ground points and non-ground points are separated through a ground point detection algorithm. The ground points are used for terrain slope calculation, and the non-ground points are used for obstacle detection; Based on the processed point cloud data, a three-dimensional voxel grid is constructed. The voxel resolution is defined as r×r×h, where r represents the horizontal resolution and h represents the vertical resolution. The global coordinate system is divided into a three-dimensional grid space V: ; Among them, represents a voxel, i and j respectively represent the indices of the voxel in each dimension on the horizontal plane in the three-dimensional grid space, k represents the index of the voxel in the vertical direction in the three-dimensional grid space, MX and MY respectively represent the number of voxels in each dimension on the horizontal plane in the three-dimensional grid space, and H represents the number of voxels in the vertical direction in the three-dimensional grid space; For any voxel , calculate the density of non-ground points in the voxel and map it to the obstacle probability through a logistic regression model , the obstacle probability . When the obstacle probability is 0, it means that there are no obstacles in the voxel. When the obstacle probability is 1, it means that the voxel is filled with obstacles and is impassable; For the voxel where the ground points fall , calculate the terrain slope through local plane fitting ; The soil compactness is obtained by collecting data from soil sensors buried in the closed park and mapping the point data to three-dimensional voxels using the inverse distance weighted interpolation method. ; For images captured by visual cameras, the obstacle categories in the images are identified through semantic segmentation algorithms, pixel-level semantic masks are generated to mark obstacle categories, and obstacles are divided into static obstacles and dynamic obstacles. The pixel coordinates of obstacles in the images are converted into global three-dimensional coordinates (X, Y, Z) through the intrinsic and extrinsic matrix of the visual camera. Construct an octree dynamic index and define octree nodes. The root node covers the entire park. Each node is divided into 8 child nodes until the minimum voxel resolution r×r×h is reached. When the visual camera recognizes a dynamic obstacle, extract the dynamic obstacle 3D bounding box, traverse the octree, mark the nodes that intersect with the bounding box as dynamic obstacle areas, and update the obstacle probability of the corresponding voxel. If it is 1, the Kalman filter is used to predict the position of the dynamic obstacle area at the next moment, and the octree node status of the next three time steps is updated in advance; Define the environment state tensor E: ; Among them, T is the time series and D is the environmental feature dimension: ; According to the spatiotemporal correlation of environmental feature dimensions, a dynamic environment model is established, and the voxel state transfer function is defined. The sensor data is synchronized in real time through the edge computing node, and the environmental state tensor is updated in each data collection cycle to ensure the consistency of the operation plan with the real-time environment. It should be noted that the role of this module is to build a high-precision dynamic environment model that includes terrain, obstacles, and soil characteristics through multi-source heterogeneous sensor data fusion, providing real-time environmental information for path planning and conflict monitoring; Path planning module: decouples complex tasks into atomic operation units, generates a directed acyclic graph of operation dependencies, uses dynamic programming to generate the optimal task sequence for a single tractor, improves the genetic algorithm to achieve global operation allocation, and uses the ant colony algorithm combined with a dynamic environment model to plan the optimal path for tractors in a closed park that takes environmental constraints into account; Decouple the complex tasks in the closed park into several minimum executable atomic work units and integrate them into a set of atomic work units. Use a multi-dimensional feature vector to describe the characteristics of each atomic work unit, including geometric features, time features, resource features, dependency features, and priority features: Among them, the geometric features are obtained by using the convex hull algorithm to describe irregular plots, the time features are determined by combining agronomic requirements and equipment scheduling constraints, the resource features include the required types of farm tools and the estimated energy consumption, the dependency features include the set of previous tasks, and the priority features are comprehensively determined by the timeliness of operations and the scarcity of resources; It should be noted that the complex task is decomposed into the smallest executable atomic job units, and the attributes of the atomic job units are described by multi-dimensional feature vectors, and the role is to provide basic units for task scheduling; Based on the multi-dimensional feature vectors of atomic job units, a directed acyclic graph of job dependencies between atomic job units is constructed. The nodes in the directed acyclic graph of job dependencies represent atomic job units, and the directed edges represent the execution order constraints of atomic job units. A time window constraint matrix is introduced to avoid time overlap between atomic job units. An initial job sequence is generated through a topological sorting algorithm to ensure that all dependency constraints are satisfied; Combined with the initial job sequence and the environmental state tensor E, a dynamic programming algorithm is used to define the state space represents the minimum cumulative cost when processing the atomic job unit with serial number a, currently using farm tool b, and the current time is t. The minimum cumulative cost is iteratively solved through the Bellman equation to generate the optimal task sequence for a single tractor; According to the optimal job sequence, the global job allocation of tractors in the closed park is realized through an improved genetic algorithm; Specifically, the improved genetic algorithm adopts a two-layer coding structure, where the upper layer is the allocation layer, and a binary matrix is defined: ; Among them, A represents the number of atomic job units, N represents the number of tractors in the closed park, represents allocating the atomic job unit with serial number a to the tractor with serial number n; The lower layer is the sequence layer, and an integer vector is defined represents the sequence of atomic job units allocated to the tractor with serial number n. The sequence of atomic job units represents the execution order of atomic job units to ensure that all dependency constraints are satisfied; In the improved genetic algorithm, a multi-objective fitness function is designed to balance efficiency, balance and safety: ; Among them, represents the time efficiency term, driving the algorithm to minimize the maximum completion time and improve the overall efficiency, represents the load balancing term, ensuring the load balance of tractors and avoiding failures caused by overwork of a certain device, represents the safety constraint term, detecting conflicts through pre-computed trajectory intersections, represents the weight calculated by the mean adaptive method; By improving the genetic algorithm and based on the optimal operation sequence, the allocation layer generates the optimal allocation matrix FP that clearly defines the tractor corresponding to each task, and the sequence layer generates the atomic operation unit sequence set for each tractor. ; Using the ant colony algorithm, according to the generated atomic operation unit sequence set and the environment state tensor E, to perform path planning for each tractor; Specifically, for the atomic operation unit sequence of the tractor with sequence number n, the geometric center coordinates of the operation area of ​​each atomic operation unit are extracted to form a node set: ; Wherein, m represents the number of atomic operation units in the atomic operation unit sequence of the tractor with sequence number n; Defining Nodes The edge weight between , integrating environmental features and geometric distance: ; Among them, r and s represent Any two different integers between represents the Euclidean distance, represents the average obstacle probability of each voxel on the path, represents the average soil compaction along the path, It represents the environmental impact coefficient, which is calibrated through closed park tests; Get the task node graph with environmental weights as the input graph structure of the ant colony algorithm; In the ant colony algorithm, pheromone initialization and ant colony behavior modeling are performed. The core parameters including the number of ants, the initial value of pheromone, the pheromone volatility rate and the heuristic factor are set. The state transition rule is defined to calculate the state transition probability of each ant selecting the next node at any node, output the initial pheromone matrix, set each ant to traverse all nodes according to the state transition probability, generate a complete path and record the total length of the path. ; ; Iterate and update the pheromone matrix to ensure that the optimal path accumulates more pheromones. When the change rate of the optimal path length for five consecutive generations is less than the change threshold, terminate the iteration and output the optimal path of the tractor with the current serial number n. It should be noted that the function of this module is to decompose, sort and allocate job tasks to multiple tractors, generate the optimal task sequences and paths for each tractor in combination with the real-time environmental status, ensure task dependency constraints and environmental safety, improve the collaborative operation efficiency, improve the double-layer coding structure and multi-objective fitness function of the genetic algorithm, comprehensively consider operation efficiency, equipment load balance and path safety, optimize the global job allocation plan. The ant colony algorithm integrates environmental features in path planning, making the planned path more in line with the actual environment and improving the feasibility and safety of the path; Conflict monitoring module: Control tractors in a closed park to perform collaborative operations through the optimal path, deploy edge computing nodes to receive tractor pose data in real time, determine the actual conflict risk by calculating the spatio-temporal conflict density within a sliding time window, and detect the predicted conflict risk by fitting the predicted trajectory function. If there is an actual conflict risk or a predicted conflict risk, trigger conflict coordination; Establish a kinematic model describing the motion state of the tractor, and control the tractors in the closed park to perform collaborative operations through the optimal path output by the path planning module; Deploy 5 edge computing nodes (ECNs) at the four corners and the center of the closed park, equipped with industrial-grade hosts, supporting wireless communication, ensuring no dead spots in the whole area, receiving the pose data of each tractor in real time. The pose data is obtained by fusing the GNSS positioning and inertial navigation results, and synchronously obtaining the real-time updated environmental state tensor E; Perform linear interpolation processing on the pose data of each tractor to generate a continuous pose sequence for each tractor; It should be noted that the edge computing node receives the tractor pose data in real time and generates a continuous pose sequence by linear interpolation. Its function is to provide high-precision trajectory data for spatio-temporal conflict analysis; Construct a dynamic three-dimensional spatio-temporal relationship matrix SK, and define a sliding time window , where TW represents the window size, t represents the current time point, calculate the Euclidean distance between two tractors with serial numbers qa and qb at each time step within the sliding time window. If the Euclidean distance is less than the safety distance: ; Otherwise: ; Among them, represents the reverse chronological serial number of any time step within the sliding time window, and qa and qb represent any two different integers between; Introduce spatio-temporal conflict density to quantify the conflict risk: ; If the spatio-temporal conflict density is greater than or equal to the conflict threshold, it is determined that there is an actual conflict risk between two tractors; Perform least squares fitting on the continuous pose sequences of each tractor within the sliding time window to obtain the predicted trajectory function of each tractor within the prediction window with the changing time as the variable, solve the intersection equation of the predicted trajectory functions between any two tractors. If the equation has positive real roots and the positive real roots are less than the duration threshold, it is determined that there is a predicted conflict risk between two tractors, and the predicted conflict time ; The prediction window starts from the current time point t and moves with the change of the current time. The window size of the prediction window is the same as that of the sliding time window; It should be noted that by calculating the Euclidean distance between tractors within the sliding time window, the actual conflict risk can be quantified by calculating the spatio-temporal conflict density. By performing least squares fitting on the continuous pose sequences and solving the intersection of the predicted trajectories, the future predicted conflict risk can be detected, realizing the dual monitoring of immediate and future conflicts; If there is no actual conflict risk and predicted conflict risk between all tractors pairwise within the closed park, it is judged that the multi-tractor collaborative operation situation within the closed park is normal; If there is an actual conflict risk or predicted conflict risk between any two tractors, it is judged that conflict coordination is required between the two tractors, and conflict coordination is triggered; It should be noted that the role of this module is to introduce the concepts of dynamic three-dimensional spatio-temporal relationship matrix and spatio-temporal conflict density, more comprehensively and quantitatively evaluate the conflict risk between tractors, combine the calculation of the intersection of predicted trajectories, discover potential conflict risks in advance, provide more sufficient time and information for conflict coordination, and use edge computing nodes for data processing and analysis, realizing real-time and efficient conflict monitoring, reducing data transmission delay and central computing pressure; Conflict coordination module: If conflict coordination is triggered, select a hierarchical coordination strategy for the tractor according to the type of conflict risk to achieve obstacle avoidance control; When the conflict monitoring module determines that there is an actual conflict risk or predicted conflict risk between two tractors with serial numbers qa and qb, conflict coordination is triggered, and obstacle avoidance control is achieved through a hierarchical coordination strategy. The hierarchical coordination strategy includes a real-time speed adjustment strategy and a local path replanning strategy; Select a coordination strategy for the two tractors according to the type of conflict risk. For the actual conflict risk, the real-time speed adjustment strategy is preferentially triggered, and then the local path replanning strategy is carried out to minimize the loss of operation efficiency after avoiding an immediate collision. For the predicted conflict risk, the local path replanning strategy is triggered to minimize the loss of operation efficiency while ensuring safety; Specifically, in the real-time speed adjustment strategy, for two tractors with serial numbers qa and qb whose detected Euclidean distance is less than the safety distance, speed decay control is implemented for the two tractors: ; where represents the safety speed decay factor, which is calibrated through experiments. represents the moving speed of the tractor with serial number qa at the current time point t. represents the moving speed of the tractor with serial number qb at the current time point t. Ensure that the moving speeds of the two tractors with serial numbers qa and qb are decayed in real time, generate temporary speed control instructions, send the calculated moving speeds of the two tractors to the underlying controllers of the corresponding tractors, and update the environmental state tensor and the tractor kinematic model simultaneously; If the moving speeds of the two tractors with serial numbers qa and qb are both less than the speed threshold at the current time point t, trigger the local path replanning strategy; If the local path replanning strategy is triggered, based on the optimal paths of the two tractors, calculate the pose data of the two tractors at the end point of the prediction window. Taking the pose data of the two tractors at the current time point as the starting point and the pose data of the two tractors at the end point of the prediction window as the end point, perform local path replanning for the two tractors through the ant colony algorithm in the path planning module to generate the local optimal paths of the two tractors; Adopt the local optimal paths of the two tractors to replace the path segments within the prediction window in the optimal paths of the two tractors, update the optimal paths of the two tractors, and obtain the updated optimal paths; Adopt the updated optimal paths to control the collaborative operation of the tractors, synchronously update the environmental state tensor and the tractor kinematic model, and realize the multi-tractor collaborative operation; It should be noted that the hierarchical coordination strategy conducts targeted processing according to the urgency and type of conflict risks. On the premise of ensuring safety, it minimizes the loss of operation efficiency to the greatest extent. The method of combining real-time speed adjustment and local path replanning can flexibly handle different types of conflicts, improve the robustness and adaptability of the system, and update the environmental state tensor and the tractor kinematic model in real time during the coordination process, ensuring the consistency and coordination of system information; The technical solution of the embodiment of the present invention is as follows: A multi-source heterogeneous sensor network is used to construct an environmental perception layer. Through spatio-temporal registration and data fusion, a three-dimensional voxel grid is constructed in the global coordinate system to generate an environmental state tensor. The sensor data is synchronized in real time to generate a high-precision dynamic environmental model. The complex task is decoupled into atomic job units to generate a job-dependent directed acyclic graph. The dynamic programming is used to generate the optimal task sequence of a single tractor, and the improved genetic algorithm is used to achieve global job allocation. Through the ant colony algorithm and combined with the dynamic environmental model, an optimal path considering environmental constraints is planned for the tractors in the closed park. The tractors in the closed park are controlled to perform collaborative operations through the optimal path. Edge computing nodes are deployed to receive the pose data of the tractors in real time. By calculating the spatio-temporal conflict density within the sliding time window, the actual conflict risk is determined, and the predicted trajectory function is fitted to detect the predicted conflict risk. If there is an actual conflict risk or a predicted conflict risk, conflict coordination is triggered. If conflict coordination is triggered, a hierarchical coordination strategy is selected for the tractors to achieve obstacle avoidance control according to the type of conflict risk.

[0014] Embodiment 2, as Figure 2 shown, the method for multi-tractor collaborative operation scheduling and control for a closed park described in the embodiment of the present invention includes the following steps: Step 1: Use a multi-source heterogeneous sensor network to generate an environmental state tensor and construct a dynamic environmental model; Step 2: Decouple the complex operation task into atomic job units, use dynamic programming to generate the optimal task sequence of a single tractor, combine the improved genetic algorithm to achieve multi-tractor global task allocation, and integrate the ant colony algorithm with the dynamic environmental model to plan the optimal path for each tractor; Step 3: Control the tractors to perform collaborative operations based on the optimal path. According to the pose data collected by the edge computing nodes in real time, analyze the spatio-temporal conflict density through the sliding time window, and combine the trajectory prediction function to detect the actual conflict risk and the predicted conflict risk, and judge whether to trigger conflict coordination; Step 4: If conflict coordination is triggered, select a hierarchical coordination strategy for the tractors to achieve obstacle avoidance control.

[0015] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art of this industry should understand that the present invention is not limited by the above embodiments. What is described in the above embodiments and the specification only illustrates the principle of the present invention. Without departing from the spirit and scope of the present invention, the present invention will have various changes and improvements, and these changes and improvements all fall within the scope of the present invention claimed. The scope of protection claimed by the present invention is defined by the appended claims and their equivalents.

Claims

1. A multi-tractor cooperative operation dispatching system for closed parks, characterized by: include: Environmental modeling module: uses a multi-source heterogeneous sensor network to generate environmental state tensors and build a dynamic environmental model; Path planning module: decouples complex tasks into atomic operation units, uses dynamic planning to generate the optimal task sequence for a single tractor, combines an improved genetic algorithm to achieve global task allocation for multiple machines, and integrates the ant colony algorithm with a dynamic environment model to plan the optimal path for each tractor; Conflict monitoring module: Controls tractors to work collaboratively based on the optimal path. According to the posture data collected in real time by the edge computing node, the spatial and temporal conflict density is analyzed through a sliding time window. The actual conflict risk and the predicted conflict risk are detected in combination with the trajectory prediction function to determine whether to trigger conflict coordination. Conflict coordination module: If conflict coordination is triggered, a hierarchical coordination strategy is selected for the tractor to achieve obstacle avoidance control.

2. The tractor multi-machine cooperative operation scheduling system for closed parks according to claim 1 is characterized by: The optimal path is planned as follows: In the ant colony algorithm, pheromone initialization and ant colony behavior modeling are performed, core parameters are set, state transition rules are defined, the state transition probability of each ant selecting the next node at any node is calculated, the pheromone matrix is ​​output, and each ant is set to traverse all nodes according to the state transition probability to generate a complete path; Obtain and calculate the total length of the complete path based on the task node graph with environmental weights, iteratively update the pheromone matrix until the optimal path is found, terminate the iteration, and output the optimal path.

3. The tractor multi-machine cooperative operation scheduling system for closed parks according to claim 2 is characterized in that: The task node graph with environmental weights is obtained as follows: The atomic operation unit sequence of the tractor is obtained, and the geometric center coordinates of the operation area of ​​each atomic operation unit are extracted to form a node set. The edge weights between nodes are obtained by integrating environmental features and geometric distance calculations. The node and edge weights are combined to obtain a task node graph with environmental weights.

4. The tractor multi-machine cooperative operation scheduling system for closed parks according to claim 3 is characterized in that: The atomic operation unit sequence is obtained as follows: The optimal operation sequence of a single tractor is obtained, and the global task allocation of multiple tractors in a closed park is realized through an improved genetic algorithm. In the improved genetic algorithm, a multi-objective fitness function is designed to balance efficiency, equilibrium and safety. A two-layer coding structure is adopted, in which the upper layer is the allocation layer, which generates the optimal allocation matrix of the tractor corresponding to each task, and the lower layer is the sequence layer, which generates the atomic operation unit sequence of each tractor.

5. The tractor multi-machine cooperative operation scheduling system for closed parks according to claim 4 is characterized in that: The optimal operation sequence of the tractor is obtained as follows: Decouple the complex tasks in the closed park into several minimum executable atomic work units, describe the characteristics of each atomic work unit through a multi-dimensional feature vector, and construct a directed acyclic graph of work dependencies between atomic work units, and generate the initial work sequence in combination with a topological sorting algorithm; Combining the initial operation sequence with the environment state tensor, a dynamic programming algorithm is used to define the state space representation of the minimum cumulative cost when processing each atomic operation unit. The minimum cumulative cost is iteratively solved through the Bellman equation to generate the optimal task sequence for a single tractor.

6. The tractor multi-machine cooperative operation scheduling system for closed parks according to claim 1 is characterized in that: The actual conflict risk is detected in the following manner: Construct a dynamic three-dimensional space-time relationship matrix, define a sliding time window, obtain the continuous posture sequence of each tractor, and calculate the Euclidean distance between any two tractors at each time step in the sliding time window. If the Euclidean distance is less than the safety distance, define the corresponding element in the dynamic three-dimensional space-time relationship matrix as 1, otherwise it is defined as 0; The spatiotemporal conflict density is calculated based on the dynamic three-dimensional spatiotemporal relationship matrix. If the spatiotemporal conflict density is greater than or equal to the conflict threshold, it is determined that there is an actual conflict risk between the two tractors.

7. The tractor multi-machine cooperative operation scheduling system for closed parks according to claim 1 is characterized in that: The continuous posture sequence of each tractor in the sliding time window is obtained for least squares fitting, and the predicted trajectory function of each tractor in the prediction window with the change time as the variable is obtained. The intersection equation of the predicted trajectory function between any two tractors is solved. If the equation has a positive real root, the positive real root represents the change time. If the change time is less than the duration threshold, it is determined that there is a prediction conflict risk between the two tractors.

8. The tractor multi-machine cooperative operation scheduling system for closed parks according to claim 6 is characterized by: The detection method of the predicted conflict risk is: A kinematic model describing the motion state of the tractor is established, and the tractors in the closed park are controlled to work in coordination through the optimal path output by the path planning module; Five edge computing nodes are deployed at the four corners and the center of the closed park. They are equipped with industrial-grade hosts and support wireless communication to ensure that there are no blind spots in the entire area. The position data of each tractor is received in real time. The position data is obtained by integrating GNSS positioning with inertial navigation results, and the real-time updated environmental state tensor is obtained synchronously. The position and posture data of each tractor are linearly interpolated to generate a continuous position and posture sequence of each tractor.

9. The tractor multi-machine cooperative operation scheduling system for closed parks according to claim 1 is characterized in that: The hierarchical coordination strategy includes a real-time speed adjustment strategy and a local path replanning strategy: If an actual conflict risk is detected, the real-time speed adjustment strategy is triggered first, followed by the local path replanning strategy. If a predicted conflict risk is detected, the local path replanning strategy is triggered; In the real-time speed adjustment strategy, speed attenuation control is implemented for two tractors with actual conflict risks. If the movement speeds of both tractors at the current time point are less than the speed threshold, the local path replanning strategy is triggered. If the local path replanning strategy is triggered, based on the optimal path of the two tractors, the posture data of the two tractors at the end of the prediction window is calculated as the end point, and the posture data of the two tractors at the current time point is used as the starting point. The local optimal path is generated for the two tractors through the ant colony algorithm, and the path segment in the prediction window in the optimal path of the two tractors is replaced. The optimal path of the two tractors is updated, and the updated optimal path is used to control the tractors to work collaboratively.

10. A method for dispatching and controlling tractor multi-machine cooperative operation in a closed park, characterized in that: The following steps are involved: Use multi-source heterogeneous sensor networks to generate environmental state tensors and build dynamic environmental models; Decouple complex tasks into atomic units, use dynamic programming to generate the optimal task sequence for a single tractor, combine improved genetic algorithms to achieve global task allocation for multiple machines, and integrate ant colony algorithms with dynamic environment models to plan the optimal path for each tractor. The tractors are controlled to work collaboratively based on the optimal path. According to the posture data collected in real time by the edge computing node, the spatiotemporal conflict density is analyzed through the sliding time window. The actual conflict risk and the predicted conflict risk are detected in combination with the trajectory prediction function to determine whether to trigger conflict coordination. If conflict coordination is triggered, a hierarchical coordination strategy is selected for the tractor to achieve obstacle avoidance control.

Images