Multi-machine collaborative operation method based on vertical domain large model

Abstract

This application relates to the field of agricultural mechanization and intelligent control technology, and discloses a method for collaborative operation of multiple agricultural machines based on a vertical domain large model. The method includes: a central controller receiving multi-source sensing data uploaded by onboard controllers, fusing and constructing a vertical domain knowledge graph and generating a load distribution prediction model, and issuing an initial collaborative path accordingly; each onboard controller collecting attitude data to calculate deviations, correcting parameters and uploading them if they exceed a threshold, and generating an optimized collaborative path; the central controller receiving torque values ​​to calculate the resultant torque distribution, determining the lever arm length distribution, analyzing the center of gravity offset in conjunction with the weight of the agricultural machinery, and issuing commands to adjust the relative posture of the agricultural machinery under high loads; each onboard controller monitoring hydraulic pressure, fine-tuning actuator resistance and uploading stability parameters if it exceeds the range; the central controller evaluating durability indicators based on historical data, and updating the collaborative path by fusing real-time data when the indicators are too low, obtaining the final solution. This application improves the accuracy of collaborative operation of multiple agricultural machines in complex terrain.

Description

Technical Field

[0001] This application relates to the field of agricultural mechanization and intelligent control technology, and in particular to a method for multi-agricultural machinery collaborative operation based on a large vertical domain model. Background Technology

[0002] In modern agricultural production, the coordinated operation of multiple agricultural machines has become an important way to improve the efficiency of field operations and reduce labor costs. However, existing coordination methods mostly rely on preset fixed paths and formations, which are difficult to adapt to the spatial heterogeneity of soil hardness and crop density within a field. When agricultural machines enter the same field, the differences in soil hardness and crop density directly affect the working load of each machine, resulting in lower resistance for machines in front and higher resistance for machines behind. Uneven load distribution leads to path tracking deviation, formation instability, continuous high-pressure output of the hydraulic system, accelerated wear of key components, and a vicious cycle.

[0003] Existing technologies lack the ability to deeply model agricultural knowledge when addressing the aforementioned problems. Traditional methods treat soil, crops, and agricultural machinery as isolated information sources, making it difficult to understand the complex relationship between soil hardness and crop density. They also fail to dynamically adjust operational strategies based on the real-time stress state of agricultural machinery, resulting in inaccurate load prediction, inflexible path planning, and untimely formation adjustments.

[0004] In recent years, the rise of vertical domain large-scale model technology has provided a new approach to solving the above problems. Based on massive amounts of agricultural data, vertical domain large-scale models can deeply understand the complex interactions between soil, crops, and agricultural machinery, and possess multi-source data fusion and knowledge reasoning capabilities. Introducing vertical domain large-scale models into multi-machine collaborative operations can construct a knowledge representation framework that characterizes the spatial heterogeneity of plots, enabling accurate prediction of load distribution, dynamic optimization of path planning, and intelligent decision-making for formation adjustments. This effectively breaks the vicious cycle of uneven load distribution, path deviation, and formation instability.

[0005] Therefore, how to construct a large-scale vertical model for agriculture in complex plots where soil hardness and crop density are spatially uneven, to achieve deep fusion of multi-source data and knowledge reasoning, to effectively perceive and respond to real-time load differences among agricultural machinery, while maintaining formation stability and ensuring long-term reliable operation of equipment, has become a key technical problem that urgently needs to be solved in this field. Summary of the Invention

[0006] To address the aforementioned technical issues, this application provides a method for multi-machine collaborative operation based on a large vertical domain model, which improves the accuracy, safety, and reliability of collaborative operation of multiple agricultural machines in complex plots.

[0007] Firstly, this application provides a method for multi-agricultural machinery collaborative operation based on a large vertical domain model, the method comprising: Step S1: The central controller receives multi-source sensing data uploaded by the airborne controller, integrates and constructs a vertical domain knowledge graph, obtains a load distribution prediction model, and generates an initial collaborative path to be sent to the airborne controllers of each agricultural machine. Step S2: The onboard controller of each agricultural machine collects its own attitude data in real time, calculates the attitude deviation from the initial collaborative path, and if it exceeds the preset threshold, it corrects the path following parameters and uploads them. The central controller then aggregates and generates an optimized collaborative path. Step S3: The central controller receives the torque values ​​collected by the load detection circuit of each agricultural machine, calculates the resultant torque distribution based on the relative positions of the agricultural machines in the optimized cooperative path, and determines the lever arm length distribution of each agricultural machine. Step S4: The central controller analyzes the center of gravity offset of the group based on the distribution of lever arm length and the weight of the agricultural machinery. If the group falls into the preset high load area, the central controller generates a formation change command and issues it. Each onboard controller responds to the command and readjusts the relative position of the machine. Step S5: Each onboard controller monitors the hydraulic pressure of the machine in real time. If the pressure reading of any agricultural machine exceeds the safe range, the actuator resistance is finely adjusted and the stable parameters are uploaded to the central controller. Step S6: The central controller combines the historical operating data and stable parameters of each agricultural machine to evaluate the durability index of each agricultural machine. If any durability index is lower than the preset threshold, the collaborative path is updated by combining real-time multi-source data fusion to obtain the final multi-agricultural machine collaborative operation scheme.

[0008] Compared with the prior art, the beneficial effects of the present invention are at least as follows: This application constructs a hierarchical collaborative architecture that includes a central controller and various agricultural machinery onboard controllers, thereby achieving efficient fusion of multi-source data and organic unity of distributed decision-making. The central controller constructs a vertical domain knowledge graph based on soil hardness and crop density data to generate a high-precision load distribution prediction model, thus avoiding the problem of uneven load caused by the spatial heterogeneity of plots from the source. Each agricultural machine collects multi-dimensional data such as attitude, torque, and hydraulic pressure in real time, and completes attitude deviation correction and hydraulic pressure closed-loop regulation at the onboard controller level to ensure the accuracy and safety of single machine operation; the central controller gathers data from each machine and optimizes the resultant torque distribution through genetic algorithms, adaptive formation transformation based on center of gravity offset, and path iterative optimization driven by durability assessment to achieve load balancing and collaborative control at the formation level. This application organically integrates the three-layer functions of perception, decision-making, and execution, forming a complete closed-loop control link from environmental modeling to path planning, from attitude correction to formation adjustment, and from pressure regulation to durability protection. This significantly improves the operating accuracy, formation stability, equipment safety, and service life of multiple agricultural machines in complex terrain, providing an innovative technical solution for the cluster collaborative operation of intelligent agricultural equipment. Attached Figure Description

[0009] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0010] Figure 1 This is a flowchart illustrating a multi-agricultural machinery collaborative operation method based on a large vertical domain model, according to an embodiment of this application. Figure 2 This is a structural diagram of a multi-machine cooperative operation device according to an embodiment of this application. Figure 3 This is a schematic diagram of path planning and restricted areas in an embodiment of this application; Figure 4 This is a schematic diagram illustrating the process of real-time monitoring and closed-loop regulation of hydraulic pressure according to an embodiment of this application. Detailed Implementation

[0011] This application provides a method for multi-machine collaborative operation based on a large vertical domain model. The terms "first," "second," "third," "fourth," etc. (if present) in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in a sequence other than that illustrated or described herein. Furthermore, the terms "comprising" or "having" and any variations thereof are intended to cover a non-exclusive inclusion; 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 explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or devices.

[0012] This application provides a method for multi-machine collaborative operation based on a large vertical domain model, such as... Figure 2 As shown, an embodiment of this application includes a multi-agricultural machinery collaborative operation method based on a large vertical domain model, comprising: Step S1: The central controller receives multi-source sensing data uploaded by the airborne controller, integrates and constructs a vertical domain knowledge graph, obtains a load distribution prediction model, and generates an initial collaborative path to be sent to the airborne controllers of each agricultural machine.

[0013] In step S1, the vertical domain knowledge graph is fused to obtain the load distribution prediction model. This includes: collecting soil hardness data and crop density image data from multiple points within the plot as multi-source sensing data; performing spatiotemporal alignment and feature-level fusion on the multi-source sensing data to obtain a fused feature vector; constructing a vertical domain knowledge graph containing plot spatial grid nodes, soil hardness attributes, crop density attributes, and their interrelationships based on the fused feature vector; deriving load influence factors for each plot grid area based on node attributes and edge relationships in the vertical domain knowledge graph; and obtaining a load distribution prediction model containing spatial heterogeneity features based on the load influence factors.

[0014] Specifically, in order to achieve accurate planning of multi-machine collaborative operation paths, it is necessary to first construct a load distribution prediction model based on the actual environmental characteristics of the plot cultivation, and then generate an initial collaborative path that adapts to the spatial heterogeneity of the plot. This avoids the problem of uneven machine load caused by differences in soil and crop distribution from the source, and ensures the rationality of field collaborative operations.

[0015] Specifically, such as Figure 2 The diagram shows the equipment structure for multi-machine collaborative operation, including a central controller and onboard controllers, positioning circuits, load detection circuits, attitude sensors, and hydraulic sensors installed on each machine. The positioning circuit collects the position data of the machines, which is the basic spatial data for determining their relative positions in the work formation and planning the collaborative operation path, providing a positional basis for subsequent path optimization and formation adjustment. The load detection circuit collects the torque data of the working parts of the machines, which is the core data reflecting the actual load during operation and is a key basis for calculating the resultant torque distribution and analyzing the load balance state. The attitude sensors collect real-time attitude data of the machines, which directly shows whether the machine's operating posture deviates from the preset path, and is important data for judging attitude deviations and correcting path following parameters. Hydraulic sensors are used to collect pressure data from the hydraulic system of agricultural machinery. This pressure data directly reflects the operating status of the hydraulic system and is the core basis for judging whether the hydraulic system is within the safe operating range and fine-tuning the actuator resistance. The onboard controller is connected to the positioning circuit, load detection circuit, attitude sensor, and hydraulic sensor. As the core of data acquisition and command execution at the agricultural machinery end, the onboard controller collects the corresponding data output by each positioning circuit, load detection circuit, attitude sensor, and hydraulic sensor in real time and uploads the collected data to the central controller, providing complete and real-time agricultural machinery operation data support for the central controller's collaborative decision-making. On the other hand, it receives various control commands issued by the central controller and strictly executes the corresponding operation parameter adjustments, formation changes, and other operations according to the commands, realizing rapid response of the agricultural machinery end to collaborative decision-making.

[0016] The central controller communicates with the onboard controllers of each agricultural machine. As the core decision-making unit for multi-machine collaborative operation, the central controller centrally receives all agricultural machine operation-related data uploaded by each onboard controller. Based on this data, it conducts collaborative decision-making throughout the entire process, including path planning, load balancing analysis, formation change decision-making, and durability assessment. At the same time, it sends the control commands generated by the decisions to the onboard controllers of each agricultural machine, realizing unified scheduling and collaborative control of multi-machine formation operation. Through the hardware configuration and data interaction logic of this system, a perception, transmission, decision-making, and execution system for multi-machine collaborative operation is established. This provides hardware support for the implementation of multi-machine collaborative operation methods based on vertical domain large models, effectively solving the problems of poor data interaction and disconnect between decision-making and execution in traditional collaborative operations, and ensuring the accurate and efficient implementation of collaborative operation control strategies.

[0017] First, the central controller receives multi-source sensing data, including soil hardness data and crop density image data. Soil hardness data is collected by an array of soil hardness sensors located in various areas of the plot. These sensors transmit real-time, multi-point soil hardness values ​​to the central controller via a wireless network. Crop density image data is captured by a high-resolution camera mounted on a drone, ensuring coverage of the entire work area. The central controller performs spatiotemporal alignment processing on the received soil hardness and crop density image data. Specifically, it aligns the timestamps of the soil hardness data with the acquisition times of the crop density image data using GPS coordinates, ensuring that both types of data are processed within the same spatiotemporal framework. After spatiotemporal alignment, the central controller extracts the numerical features of the soil hardness data and the pixel density features of the crop density image, and performs feature-level fusion using a weighted average method to obtain a fused feature vector. The weighting coefficients are calculated based on data reliability, with the weights for soil hardness data determined by sensor accuracy error and the weights for crop density images determined by image resolution.

[0018] Subsequently, the central controller constructs a vertical domain knowledge graph based on the fused feature vectors. First, the land parcels are divided into several grid nodes, each corresponding to a unique geospatial location. Soil hardness and crop density attribute values ​​are then assigned to each node. The soil hardness attribute value comes from the hardness component of the fused feature vector, and the crop density attribute value comes from the density component of the fused feature vector. After assigning node attributes, the central controller establishes edge relationships between nodes to characterize the degree of association between different grid regions. For any two grid nodes i and j, the central controller extracts their soil hardness and crop density attribute values ​​to form corresponding attribute vectors and calculates the normalized Euclidean distance between the two attribute vectors. Then, attribute similarity is calculated based on Euclidean distance. The calculation formula is: The attribute similarity value ranges from 0 to 1, with values ​​closer to 1 indicating greater similarity in attributes between two nodes. The central controller presets an attribute similarity threshold, for example, 0.8. For all grid node pairs, if the calculated attribute similarity is greater than or equal to this preset threshold, an association edge is established between node i and node j, and the attribute similarity value is stored as the edge weight. In this way, not only spatially adjacent nodes can be connected, but spatially non-adjacent nodes with similar attribute characteristics can also establish logical associations, thus forming an edge relationship network in the vertical domain knowledge graph that reflects the spatial heterogeneity of land parcels. As an example, in a cornfield application scenario, although the northeast and southwest regions of the plot are far apart, their soil hardness values ​​are both in the range of 40-45 kPa, and their crop density values ​​are both in the range of 25-30 plants per square meter. The calculated attribute similarity reaches 0.85, which exceeds the preset threshold of 0.8. Therefore, an association edge is established between the grid nodes of the two regions. This association edge allows the load characteristics of the two regions to corroborate each other when the load influence factor is subsequently deduced, thereby improving the coverage integrity of the load distribution prediction model for spatially heterogeneous regions.

[0019] After constructing the edge relationships, the central controller obtains a complete vertical domain knowledge graph, which is stored in a graph database in graph structure form, containing plot grid nodes, node attributes, and the associated edges and edge weights between nodes. Based on the node attributes and edge relationships in the vertical domain knowledge graph, the central controller deduces the load influence factors for each plot grid area. Specifically, a graph traversal algorithm is used to traverse and calculate the vertical domain knowledge graph. In this embodiment, a breadth-first search algorithm is used, starting from any node in the graph and visiting all reachable nodes along the associated edges. During the traversal, for each grid node k, the central controller considers the node's own soil hardness attributes... With crop density attributes Calculate the load influence factor of a node by considering the attributes and edge weights of its connected neighboring nodes. The calculation formula is: ,in, Let k be the set of neighboring nodes connected to node k via an associated edge. The weight of the edge connecting node k and its neighbor node j, i.e., the attribute similarity. , The number of neighboring nodes. , and Preset weighting coefficients are used to adjust the contribution of the self-attribute and neighbor attributes to the load influence factor, as in this embodiment. Take 0.5, Take 0.3, Let's take 0.2. This formula consists of three terms, the first term... The second term reflects the contribution of the soil hardness at the node to the load; the higher the soil hardness, the greater the resistance. The third term reflects the contribution of the node's own crop density to the load; the higher the crop density, the greater the drag. The load factor reflects the corroborating effect of neighboring regions similar to a node on the load. By incorporating the attribute information of neighboring nodes, the load influence factor can reflect the spatial clustering characteristics of soil hardness and crop density within a local area. In actual farmland, soil hardness and crop density often exhibit regional clustering characteristics. Hard soil areas are usually contiguous, and densely planted areas are also often distributed in patches. If only the attributes of a single node are considered, misjudgments may occur due to sensor measurement errors or local anomalies. Introducing the neighbor term achieves three effects: First, it smooths outliers. If the hardness sensor reading of a node is abnormally high, but the hardness of the surrounding similar areas is normal, the neighbor term will lower the load factor of that node. Second, it enhances regional characteristics. If an area is entirely hard soil, the nodes corroborate each other through associated edges, and the load factor increases accordingly. Third, it reflects spatial heterogeneity. Truly high-load areas will form clusters of high-load nodes in the map rather than isolated high-value points.

[0020] After the traversal calculation is completed, the central controller obtains the load influence factor distribution covering all grid nodes of the entire plot. For isolated nodes not visited during the traversal, the central controller directly calculates the load influence factor based on its own attributes, that is, by using the method of only retaining its own attribute terms in the above formula, or by supplementing it with the calculation results of neighboring visited nodes through Kriging interpolation. Based on the derived load influence factors, the central controller generates a load distribution prediction model that includes spatial heterogeneity characteristics. This model is stored in the form of a load influence factor distribution matrix, where the row and column indices of the matrix correspond to the spatial location of the plot grid, and the matrix element values ​​are the load influence factor values ​​of the corresponding grid nodes. In essence, this constitutes a high-precision load map reflecting the operational resistance level of each area of ​​the plot.

[0021] The above steps, by constructing a vertical domain knowledge graph and deriving load influencing factors, integrate discrete soil hardness and crop density data into a continuous high-precision load map, thereby realizing the quantitative expression of plot spatial heterogeneity.

[0022] In step S1, generating an initial collaborative path includes: extracting areas where soil hardness is higher than a preset hardness threshold and areas where crop density is higher than a preset density threshold from the load distribution prediction model as comprehensive restricted areas; dividing the target operation area into grids and marking the comprehensive restricted areas as insurmountable obstacles; and using a path planning algorithm, combined with a vertical domain knowledge graph, to generate an initial collaborative path that bypasses the comprehensive restricted areas and covers the target operation area.

[0023] Specifically, the central controller generates an initial collaborative path based on the load distribution prediction model. The load distribution prediction model is stored in the form of a load influence factor distribution matrix. The matrix element values ​​correspond to the load influence factor values ​​of each grid node. Areas with higher load influence factors indicate higher soil hardness or higher crop density, and greater resistance that agricultural machinery needs to overcome during operation.

[0024] The central controller first extracts restricted areas from the load distribution prediction model. Specifically, the central controller traverses all grid nodes in the load distribution prediction model, marking nodes with soil hardness attribute values ​​exceeding a preset hardness threshold as hardness-restricted nodes and nodes with crop density attribute values ​​exceeding a preset density threshold as density-restricted nodes. The preset hardness and density thresholds are pre-set based on the operational scenario and type of agricultural machinery. For example, for a rice paddy scenario, the preset hardness threshold is set to 40 kPa and the preset density threshold to 15 plants per square meter; for a cornfield scenario, the preset hardness threshold is set to 60 kPa and the preset density threshold to 10 plants per square meter. The central controller overlays all grid nodes marked as hardness-restricted and density-restricted nodes to form a comprehensive restricted node set, and connects the grid regions corresponding to these nodes into a continuous region to obtain the comprehensive restricted area. The comprehensive restricted area is the unoperable area that the initial collaborative path needs to avoid.

[0025] After determining the comprehensive restricted area, the central controller divides the target work area into a grid consistent with the load distribution prediction model, and marks all grids within the comprehensive restricted area as insurmountable obstacles. The boundary of the target work area is pre-determined based on the plot size and work tasks to ensure coverage of all farmland requiring work. Subsequently, the central controller uses a path planning algorithm to generate an initial collaborative path. In this embodiment, the A* algorithm is used as the path planning algorithm. This algorithm finds the optimal path by calculating the cost function from the starting point to the target point. The specific process is as follows: First, constructing the search space for path planning: The central controller uses the grid map of the target work area as the search space for the A* algorithm, where each grid is a path node. Each node is associated with a load influence factor obtained from the load distribution prediction model, which quantifies the expected operational resistance when the agricultural machinery passes through that grid. Second, defining the passage cost between nodes: In the A* algorithm, the cost of moving from the current node to an adjacent node is crucial in determining the path direction. The central controller defines the movement cost function in conjunction with the vertical domain knowledge graph. This function represents the function from the node Move to its neighboring node The required cost is calculated using the following formula: ,in, For nodes Move to its neighboring node The physical distance; It comes directly from the node. Load Influence Factor Load Influence Factor The higher the value, the greater the operational resistance in that area, and the higher the cost of movement. This also increases, meaning that the A* algorithm tends to choose [a specific path] when pathfinding. Smaller low-load areas allow for a strategy that prioritizes low-load areas.

[0026] Next, access barriers are set for restricted areas. For grid nodes marked as comprehensively restricted areas, the central controller sets their movement cost to infinity. In the algorithm logic, an infinite cost means that the node is impassable, thus ensuring that the planned path physically bypasses all restricted areas. Then, a heuristic function is introduced to guide the search direction; the efficiency of the A* algorithm depends on this heuristic function. This function estimates the path cost from the current node n to the target node. In this embodiment, Euclidean distance or Manhattan distance is used as a heuristic function to guide the search direction towards the destination, ensuring that the algorithm does not deviate from the target area while searching for detour paths. Finally, iterative optimization and path generation are performed. The A* algorithm starts from the starting node and comprehensively considers the actual cost from the starting node to the current node. The estimated cost from the current node to the destination The algorithm considers the total cost of each node and repeatedly explores the node with the minimum total cost using a priority queue, gradually expanding the search outwards until the endpoint is found. When the algorithm terminates, the sequence of nodes traversed from the endpoint back to the starting point constitutes the final generated path. Through this mechanism, the central controller generates a complete initial collaborative path. This path macroscopically covers the entire target work area, microscopically bypasses all comprehensive restricted areas, and prioritizes routes with low load impact factors within the passable area, achieving the dual goals of work efficiency and equipment protection.

[0027] like Figure 3As shown in the diagram, the path planning and restricted area illustration in this embodiment visually demonstrates the spatial layout of the path planning process described above. The plot is divided into a 10x12 grid background, with each grid representing a 0.8m x 0.8m actual work area. Three types of restricted areas are marked with different filling patterns: the right-diagonal filled area in the upper right region is the hardness restricted area, representing areas where soil hardness exceeds a preset threshold; the left-diagonal filled area in the lower left region is the density restricted area, representing areas where crop density exceeds a preset threshold; and the grid-line filled area in the middle region is the comprehensive restricted area, representing areas that simultaneously meet both hardness and density threshold conditions. The passable areas are the blank areas in the grid background, allowing agricultural machinery to travel in convoys. The starting point S is located in the lower left grid and is marked with a white dot; the ending point E is located in the upper right grid and is marked with a dark gray dot. The black solid arrow from the starting point S to the ending point E represents the initial collaborative path, which prioritizes passable areas with lower load impact factors, reflecting the path planning algorithm's comprehensive consideration of work efficiency and equipment protection. Based on the initial collaborative path, three parallel trajectories are drawn in the figure using dark gray dashed lines, labeled as Agricultural Machinery 1, Agricultural Machinery 2, and Agricultural Machinery 3, respectively. The trajectory of Agricultural Machinery 2 coincides with the initial collaborative path, and the trajectory spacing is marked as 2 meters to ensure that multiple agricultural machinery formations do not interfere with each other during collaborative operations and that the coverage areas do not overlap or omission. This schematic diagram clearly shows the spatial concepts of identifying restricted areas, path detour strategies, and multi-machine trajectory allocation, providing intuitive visual support for understanding the path planning process in step S1.

[0028] Step S2: The onboard controller of each agricultural machine collects its own attitude data in real time, calculates the attitude deviation from the initial collaborative path, and if it exceeds the preset threshold, corrects the path following parameters and uploads them. The central controller then aggregates and generates an optimized collaborative path.

[0029] Step S2 includes: the onboard controller of each agricultural machine collects the attitude data of each agricultural machine in real time through the attitude sensor installed on the machine; calculates the deviation vector between the actual attitude of each agricultural machine and the given attitude in the initial cooperative path; calculates the magnitude of the deviation vector and determines whether the magnitude exceeds the preset attitude deviation threshold. If it exceeds the threshold, the proportional-integral-derivative control algorithm is used to adjust the path following parameters of the machine to obtain the corrected path following parameters; the onboard controller of each agricultural machine uploads the corrected path following parameters to the central controller, which then aggregates and generates an optimized cooperative path based on the current position of each agricultural machine and the corrected parameters.

[0030] Specifically, as the agricultural machinery convoy travels along the initial coordinated path, the actual driving posture of each agricultural machine will deviate from the expected posture of the initial path planning due to factors such as changes in soil hardness, terrain undulations, or uneven crop distribution within the plot. In order to ensure that the convoy can accurately follow the predetermined trajectory, it is necessary to detect and correct these posture deviations in real time.

[0031] Each agricultural machine's onboard controller collects the machine's attitude data in real time via attitude sensors mounted on the machine. These attitude sensors include gyroscopes and accelerometers, and collect the machine's yaw, pitch, and roll angles at a sampling frequency of once per second, which are recorded as follows: , and These three angles constitute a three-dimensional vector describing the current spatial attitude of the agricultural machinery. Simultaneously, the onboard controller obtains the desired attitude of the agricultural machinery at the current moment from the initial cooperative path issued by the central controller. The desired attitude includes the desired yaw angle. Desired pitch angle and expected roll angle This forms the desired attitude vector. The airborne controller compares the actual attitude vector with the desired attitude vector and calculates the deviation vector between them. This deviation vector reflects the degree to which the agricultural machinery deviates from the desired path in its three rotational degrees of freedom. To quantify the overall deviation, the onboard controller calculates the magnitude of the deviation vector. This magnitude is a scalar representing the overall degree of deviation between the agricultural machinery's attitude and the desired attitude. The calculated deviation magnitude is compared with a preset attitude deviation threshold, which is pre-set according to the operational accuracy requirements. For example, it can be set to 3 degrees for high-precision operation scenarios and 5 degrees for ordinary operation scenarios. If the deviation magnitude does not exceed the threshold, it means that the current following accuracy of the agricultural machinery meets the requirements and no adjustment is needed; if the deviation magnitude exceeds the threshold, it means that the agricultural machinery deviates significantly and a correction process needs to be initiated.

[0032] When the magnitude of the attitude deviation exceeds a preset threshold, the airborne controller initiates a proportional-integral-derivative (PID) control algorithm to dynamically adjust the path-following parameters. The PID algorithm is a closed-loop control method based on error feedback. Its input is the attitude deviation magnitude, and its output is the correction value for the path-following parameters, specifically including adjustments to the path-following gain coefficient and the aiming distance. The airborne controller first calculates the current attitude deviation magnitude, which serves as the input to the PID controller, and the output of the PID controller... It is composed of three superimposed terms, and the calculation formula is: The first item is the proportional control item. This is a proportionality coefficient, and this term is based on the magnitude of the deviation at the current time t. The output is proportional to the magnitude of the deviation; the greater the deviation, the larger the output, enabling the agricultural machinery to respond instantly to path deviations at the current moment. The second term is the integral control term. The first term is the integral coefficient, which accumulates historical deviations to eliminate steady-state errors caused by factors such as system friction or terrain changes, preventing agricultural machinery from maintaining small deviations for extended periods and failing to fully return to the desired path; the second term is the differential control term. The differential coefficient is based on the rate of change of the deviation. Predicting the trend of deviation and increasing the adjustment amount in advance when the deviation increases rapidly enables agricultural machinery to respond proactively to changes in the path ahead, avoiding over-adjustment or delayed adjustment. It is a comprehensive control signal whose magnitude reflects the overall adjustment intensity required to eliminate the current attitude deviation. The control quantity itself is a dimensionless value. The airborne controller maps the output of the proportional-integral-derivative controller to a specific path following parameter correction value.

[0033] Specifically, the correction amount of the path following gain coefficient. With output Proportional, that is ,in, For gain mapping coefficients, the corrected path follows the gain coefficients. Correction amount for aiming distance Also with output Proportional, that is ,in This refers to the aiming distance mapping coefficient, which represents the corrected aiming distance. In this way, the output of the proportional-integral-derivative (PID) controller simultaneously affects two key parameters, enabling the path-following system to adjust both response sensitivity via the gain coefficient and forward capability via the aiming distance. This adjustment process continues in discrete time cycles. Within each control cycle, the onboard controller re-acquires attitude data, calculates a new deviation magnitude, inputs it into the PID controller to calculate a new output, and updates the path-following parameters accordingly. As the deviation magnitude gradually decreases, the output of the PID controller also decreases, forming a negative feedback loop. When the deviation magnitude converges to within a preset threshold, the PID controller stops actively adjusting, and the path-following parameters remain at their current values ​​until the next deviation exceedance event occurs. The onboard controller uses the final path-following gain coefficient and aiming distance parameters obtained after PID adjustment as the corrected path-following parameters, and uploads these corrected parameters, along with the machine's current position coordinates and attitude data, to the central controller. The uploaded data includes the machine's identifier, the current timestamp, real-time position coordinates, real-time attitude data, and the corrected path-following gain coefficient and aiming distance parameters.

[0034] After receiving the corrected parameters and real-time pose information uploaded by all online agricultural machinery, the central controller initiates the convergence and generation process for the optimized collaborative path. First, based on the optimized collaborative path at the current moment, the central controller uses the current position uploaded by each agricultural machinery as a reference point for the actual position of the formation. Since each agricultural machinery has independently corrected its parameters according to its own attitude deviation, the actual driving state of each machinery may differ. The central controller needs to incorporate these differences into the overall path optimization consideration. The central controller analyzes the corrected path following parameters uploaded by each agricultural machinery, identifying agricultural machinery with large parameter adjustment ranges and high deviation areas causing parameter adjustments. For agricultural machinery with parameter adjustment ranges significantly exceeding the average level, the central controller determines that the area where the agricultural machinery is located may have special terrain or crop conditions, and will pay special attention to this area in subsequent path planning. Based on the real-time position coordinates of all agricultural machinery, a spatial distribution map of the multi-agricultural machinery formation at the current moment is constructed. The position points of each agricultural machinery are connected according to the formation order to obtain a discrete sampling point sequence of the actual driving trajectory. The central controller compares these actual position points with the expected position points on the current optimized collaborative path to calculate the overall path following deviation distribution of the formation.

[0035] Based on this, the central controller regenerates the optimized cooperative path using curve fitting or interpolation algorithms. Specifically, the central controller uses the current position of each agricultural machine as the control points that the path must pass through. Combining this with the corrected pre-aiming distance parameters uploaded by each agricultural machine, it uses cubic spline interpolation or Bézier curve fitting methods to generate a smooth, continuous new optimized cooperative path that passes through all control points. During curve generation, the central controller adjusts the radius of curvature of the curve according to the corrected pre-aiming distance parameters of each agricultural machine. The curve curvature of the area corresponding to the agricultural machine with a larger pre-aiming distance is appropriately reduced to make the path smoother and adapt to its forward-looking requirements; the curve curvature of the area corresponding to the agricultural machine with a smaller pre-aiming distance is appropriately increased to make the path more closely match its real-time response characteristics. The newly generated optimized cooperative path simultaneously meets the following constraints: the path is continuous and smooth with no abrupt changes; the path passes through the current position of each agricultural machine; the overall trend of the path is consistent with the initial cooperative path; the path avoids all known comprehensive restricted areas; and the path prioritizes areas with low load influence factors within the passable area. The central controller stores the newly generated optimized cooperative path in the form of a path point sequence, with each path point containing information such as spatial coordinates, desired travel speed, and desired attitude angle. After the optimized collaborative path is generated, the central controller sends the path to the onboard controllers of each agricultural machine, which will serve as the collaborative path benchmark for each agricultural machine to follow in the next stage.

[0036] Through the aforementioned convergence and generation mechanism, the central controller achieves a closed loop from distributed correction to centralized optimization. The local corrections made by each agricultural machine based on its own perception are aggregated and integrated by the central controller, transforming them into a global optimization of the entire formation path. This ensures that the final optimized collaborative path can adapt to local terrain changes while maintaining the overall collaborative consistency of the formation.

[0037] Step S3: The central controller receives the torque values ​​collected by the load detection circuit of each agricultural machine, calculates the resultant torque distribution based on the relative positions of the agricultural machines in the optimized cooperative path, and determines the lever arm length distribution of each agricultural machine.

[0038] Step S3 includes: each agricultural machine, through a load detection circuit and a torque sensor installed on its tillage device, collects the torque value of its working parts in real time and uploads the torque value to the central controller; the central controller extracts the relative position coordinates of each agricultural machine in the formation from the optimized collaborative path and constructs a position relationship matrix including lateral spacing and longitudinal offset; using the torque value and position relationship matrix as input, a genetic algorithm is used to iteratively optimize and calculate the resultant torque distribution formed by the superposition of multiple machine loads; and the lever arm length distribution corresponding to each agricultural machine in the formation is determined based on the resultant torque distribution.

[0039] Specifically, during the operation of agricultural machinery convoy along the optimized collaborative path, the real-time torque borne by each working component of the agricultural machinery can reflect the immediate effect of soil resistance on the agricultural machinery. The central controller receives and analyzes these torque data, combines them with the relative positional relationship between the agricultural machinery, calculates the overall resultant torque distribution of the convoy, and then determines the lever arm length distribution of each agricultural machinery, providing a basis for subsequent formation adjustment and load balancing.

[0040] Specifically, the load detection circuit of each agricultural machine collects the torque value of the machine's working parts in real time through torque sensors installed on the agricultural machine's tillage device. The torque sensors are installed at the connection of the drive shaft or suspension device of the agricultural machine's tillage blades, and collect torque data at a preset sampling frequency, such as once per second. The collected torque values ​​are converted into digital signals and transmitted to the machine's onboard controller. The onboard controller adds a timestamp and agricultural machine identifier to each torque value, forming a torque data packet with a spatiotemporal tag. The torque data packet is then uploaded to the central controller through a wireless communication module.

[0041] After receiving torque data packets uploaded by all online agricultural machinery, the central controller first performs time alignment on each group of data to ensure that the torque values ​​used for calculation come from the same time window. For each time window, the central controller extracts the torque acquisition value of each agricultural machinery. Subsequently, the central controller extracts the relative position coordinates of each agricultural machinery in the formation from the optimized collaborative path at the current moment. The optimized collaborative path is stored in the form of a path point sequence, with each path point containing spatial coordinates and expected passage time. Based on the current position and expected passage time of each agricultural machinery, the central controller determines the relative position of each agricultural machinery in the formation, including the lateral spacing and longitudinal offset between the leading and following agricultural machinery. These relative position parameters are organized into a position relationship matrix R, where each row of the matrix corresponds to one agricultural machinery, and each column represents the lateral and longitudinal coordinates of the agricultural machinery relative to the formation reference point.

[0042] Next, the central controller uses the torque acquisition values ​​and the position relationship matrix R as input to the genetic algorithm, and iteratively optimizes the resultant torque distribution formed by the superposition of multiple machine loads using the genetic algorithm. The genetic algorithm is an optimization method that simulates the natural selection process, iteratively finding the optimal solution through operations such as encoding, selection, crossover, and mutation. In this embodiment, the specific implementation process of the genetic algorithm is as follows: The central controller first initializes the population, using the relative position offset of each machine in the formation as an individual code. Each individual consists of a set of position adjustment coefficients, representing the lateral offset and longitudinal misalignment of each machine relative to the current optimized collaborative path. Several individuals are randomly generated in the initial population, and the position adjustment coefficients of each individual are randomly assigned within a preset range. Secondly, the central controller defines the fitness function as maximizing the balance of the resultant torque after the superposition of multiple machine loads, i.e., minimizing the variance of the resultant torque distribution. For each individual in the population, the central controller corrects the position coordinates of each machine according to its position adjustment coefficients. Then, based on the corrected position coordinates and real-time torque acquisition values, the resultant torque distribution is calculated through vector superposition. The calculation formula is: ,in, Let be the torque value collected for the i-th agricultural machine. Let i be the position vector of the i-th agricultural machine. Then calculate the variance of the resultant torque as the fitness evaluation value. The smaller the fitness function value, the more balanced the force on each agricultural machine is caused by the position adjustment scheme corresponding to that individual.

[0043] Subsequently, the central controller iteratively optimizes the population. In each generation, superior individuals are selected to advance to the next generation based on their fitness function values. Assuming a selection probability of 0.8, individuals with higher fitness have a higher probability of being selected. A crossover operation is performed on the selected individuals, with a crossover probability of 0.6, randomly swapping some weight vectors of two individuals to generate new offspring. A mutation operation is then performed on the offspring individuals, with a mutation probability of 0.01, randomly adjusting the values ​​of some weight coefficients to increase population diversity. The iteration count is preset to 100 generations. After each iteration, the fitness function value of the new population is recalculated. Finally, the central controller determines whether the iteration has converged. The iteration process terminates when the change in the optimal fitness function value over several consecutive generations is less than a preset threshold, or when the maximum number of iterations (100 generations) is reached. At this point, the individual with the smallest fitness function value is the optimal solution, and its corresponding position adjustment coefficient ensures that the relative positions of the agricultural machines in the formation are optimally configured, thus achieving the most balanced distribution of the resultant torque after the multi-machine load is superimposed.

[0044] The central controller calculates the optimized resultant torque distribution based on the optimal individual position adjustment coefficients. This distribution, represented as a vector, includes both the magnitude and direction of the resultant torque, reflecting the spatial equilibrium of the overall load in the formation. Subsequently, the central controller determines the lever arm length distribution for each agricultural machine within the formation based on the resultant torque distribution. Specifically, the resultant torque component for each agricultural machine is extracted from the resultant torque distribution. Simultaneously calculate the total resultant force on the formation. The total resultant force is obtained by vector synthesis of the torque values ​​of all agricultural machinery; and according to the lever arm calculation formula: Calculate the lever arm length of each agricultural machine i. The calculated lever arm length distribution is an array corresponding to the agricultural machinery number, and each element represents the distance of the corresponding agricultural machinery relative to the center of force of the entire formation.

[0045] By optimizing the resultant torque distribution through genetic algorithms, the stress state of each agricultural machine is accurately quantified, and the abstract load perception is transformed into a specific lever arm length distribution, providing a scientific basis for subsequent adaptive adjustment of formation and significantly improving the load balance and system stability of multi-machine collaborative operation.

[0046] Step S4: The central controller analyzes the center of gravity offset of the group based on the distribution of lever arm length and the weight of the agricultural machinery. If it falls into the preset high load area, it generates a formation change command and issues it. Each onboard controller responds to the command and readjusts the relative position and posture of the machine.

[0047] Step S4 includes: the central controller calculates the overall center of gravity position of the multi-machine group based on the lever arm length distribution and the mass parameters of each machine; the overall center of gravity position is compared with the initial reference center of gravity position to calculate the group center of gravity offset; it is determined whether the group center of gravity offset falls into the preset high load area. If it does, the central controller generates a formation transformation command including lateral spacing adjustment and longitudinal misalignment adjustment, and sends it to the onboard controller of each machine; the onboard controller of each machine responds to the command, performs formation adaptive transformation, adjusts the lateral spacing and longitudinal misalignment of the machine in the multi-machine formation, and redistributes the relative posture.

[0048] Specifically, the central controller analyzes the overall center of gravity position of the multi-machine group based on the lever arm length distribution and the mass parameters of each machine. The lever arm length distribution is determined by the resultant torque distribution calculated by the genetic algorithm in the previous step, reflecting the distance of each machine relative to the overall force center of the formation. The mass parameters of each machine are known fixed values ​​and are pre-stored in the central controller's storage module, including the total mass, load mass, and operating load mass of each machine.

[0049] When calculating the overall center of gravity position of a multi-machine group, the central controller uses a weighted average algorithm to extract the mass parameters of each machine i in the group. Current position coordinates and lever arm length ,in, The lever arm length is obtained from the optimized collaborative path, and the overall center of gravity coordinates are obtained from the lever arm length distribution. This formula introduces the lever arm length as a weighting factor, allowing agricultural machinery under greater force to contribute more to the center of gravity position, thus more accurately reflecting the dynamic center of gravity of the formation under actual stress. After calculation, the central controller obtains the overall center of gravity position of the formation at the current moment. Once the overall center of gravity position is obtained, the central controller compares it with the initial reference center of gravity position to calculate the group's center of gravity offset. The initial reference center of gravity position is the center of gravity position of the formation under ideal conditions without load, and can be predetermined through static measurement or theoretical calculation before the formation enters the work area. The center of gravity offset is the vector difference between the overall center of gravity position and the initial reference center of gravity position, including the offset direction and offset distance. This offset directly reflects the degree of center of gravity change caused by uneven load during actual operation.

[0050] The central controller maps the current overall center of gravity position of the formation to the plot grid, queries the load influence factor corresponding to that position, and if the load influence factor of that position exceeds the preset load threshold, and the magnitude of the center of gravity offset exceeds the preset offset threshold, then it is determined that the group's center of gravity offset falls into the preset high load area. This judgment condition takes into account both the absolute load level of the center of gravity position and the degree of offset relative to the initial position, ensuring that the formation adjustment is triggered accurately and in a timely manner.

[0051] Once it is determined that the group's center of gravity offset falls into the preset high-load area, the central controller immediately initiates the formation adaptive transformation process. The goal of the formation adjustment is to change the lateral spacing and longitudinal misalignment between the agricultural machines, so that the adjusted load center of gravity position returns to or approaches the geometric center of the formation, thereby achieving load balance. The adjustment amount is calculated based on the vector characteristics of the center of gravity offset and quantified using a proportional control principle. Specifically, assuming the current formation is a straight formation with each agricultural machine arranged at equal lateral spacing, the central controller acquires the center of gravity offset. This offset is a vector and includes a lateral offset component. and longitudinal offset components The lateral offset component represents the distance the center of gravity deviates from the formation centerline in the left-right direction, while the longitudinal offset component represents the distance the center of gravity deviates from the formation centerline in the forward direction. Lateral spacing adjustment amount. The calculation formula is: ,in, This is a lateral adjustment coefficient, which is pre-calibrated based on the type of agricultural machinery, the working width, and soil conditions. For the lateral offset component of the center of gravity, when A positive value indicates that when the center of gravity shifts to the right, the central controller increases the lateral distance between the right-side farm machinery and the adjacent farm machinery. A negative value indicates that when the center of gravity shifts to the left, the lateral distance between the left-side farm machinery and adjacent farm machinery is increased. By increasing the lateral distance on the offset side, the farm machinery on that side gains more working space, thereby reducing its load concentration. Longitudinal misalignment adjustment amount. The calculation formula is: ,in, This is the longitudinal adjustment factor, which is pre-calibrated based on the dynamic characteristics of agricultural machinery, operating speed, and soil conditions. For the longitudinal offset component of the center of gravity, when A positive value indicates that when the center of gravity shifts forward, the central controller increases the longitudinal misalignment between the leading and following agricultural machinery, causing the leading machinery to move further forward; when... A negative value indicates that when the center of gravity shifts backward, the lag of the following agricultural machinery is increased. By adjusting the longitudinal misalignment, the time difference between each agricultural machine entering the high-load area is changed, thus dispersing the load in the time dimension.

[0052] The above adjustment coefficients are determined based on the formation dynamics and operational experience. For example, for a formation of three agricultural machines, if the center of gravity shifts 0.5 meters to the right, take... =0.3, then the horizontal spacing adjustment amount =0.15 meters, meaning the distance between the right-hand farm machinery and the middle farm machinery increases by 0.15 meters; if the center of gravity shifts forward by 0.3 meters, take The vertical misalignment adjustment amount =0.06 meters, meaning the leading agricultural machine is 0.06 meters ahead of the middle agricultural machine. For non-linear formations such as V-shaped formations, the adjustment calculation needs to consider the geometric relationship of the formation. The central controller first projects the center of gravity offset onto the horizontal and vertical axes of the formation coordinate system, and then allocates the adjustment amount according to the role of each agricultural machine in the formation. For example, in a V-shaped formation, the lateral spacing adjustment of the agricultural machines on both sides needs to be symmetrically allocated to maintain the symmetry of the formation. The central controller superimposes the calculated lateral spacing adjustment and vertical misalignment adjustment amount with the current formation parameters to obtain the adjusted target lateral spacing and target vertical misalignment amount. This, along with the number of the agricultural machine performing the adjustment, is encapsulated into a formation transformation command. After the command is issued, the onboard controller of each agricultural machine performs posture adjustment according to the target parameters, so that the entire formation reaches a new equilibrium state.

[0053] The above steps transform the abstract center of gravity offset into specific lateral spacing and longitudinal misalignment adjustment amounts. Through proportional control, adaptive formation transformation is achieved, allowing the load center of gravity to return to the geometric center of the formation, significantly improving the load balance and formation stability of multi-aircraft collaborative operations.

[0054] Step S5: Each onboard controller monitors the hydraulic pressure of the machine in real time. If the pressure reading of any agricultural machine exceeds the safe range, the actuator resistance is finely adjusted, and the stable parameters are uploaded to the central controller.

[0055] Step S5 includes: S51: The onboard controller of each agricultural machine collects real-time pressure readings of the machine's hydraulic system at a preset sampling frequency through pressure sensors installed on the hydraulic pipeline, forming a pressure reading sequence; S52: The current pressure reading in the pressure reading sequence is compared with the preset upper and lower safe pressure limits to determine whether it exceeds the safe range. If it does, the pressure error value between the pressure reading and the median value of the safe range is calculated; S53: Using a proportional-integral-derivative control algorithm, the adjustment amount of the actuator resistance is calculated based on the pressure error value, and the output resistance of the corresponding actuator of the machine is finely adjusted in a closed loop; S54: Steps S51 to S53 are repeated until the current pressure reading stabilizes within the preset safe range, obtaining stable parameters; S55: The onboard controller of each agricultural machine uploads the stable parameters to the central controller.

[0056] Specifically, after the agricultural machinery convoy completes its adaptive formation transformation and achieves a balanced load distribution, each machine enters a stable operating phase. However, due to localized abrupt changes in soil hardness or instantaneous variations in crop density, the hydraulic system pressure of each machine may still fluctuate. To ensure continuous operation within a safe load range, real-time monitoring and closed-loop regulation of the hydraulic pressure are necessary. Specifically, such as... Figure 4 The diagram shows a process flow chart for real-time monitoring and closed-loop regulation of hydraulic pressure. First, the onboard controller of each agricultural machine collects the pressure readings of its hydraulic system in real time through pressure sensors installed on the hydraulic pipeline at a preset sampling frequency. The pressure sensors are installed at the hydraulic pump outlet, actuator inlet, or key hydraulic branch. The sampling frequency is set according to the system response requirements. In this embodiment, it is set to once per second. The onboard controller adds a timestamp to each collected pressure reading and stores it in the local cache, forming a pressure reading sequence arranged in chronological order. This sequence records the dynamic process of hydraulic system pressure changing over time.

[0057] The onboard controller extracts the latest pressure reading from the pressure reading sequence and compares it with preset upper and lower safety pressure limits. These limits are predetermined based on the hydraulic system's design parameters and the safety requirements of agricultural machinery operations. For example, the upper limit might be set at 5000 Pascals and the lower limit at 2000 Pascals, representing the highest pressure the hydraulic system can withstand and the lowest pressure required for normal operation, respectively. If the current pressure reading is between the upper and lower limits, the hydraulic system is operating within a safe range, and the onboard controller continues monitoring the pressure reading for the next moment. If the current pressure reading is higher than the upper limit or lower than the lower limit, the pressure is determined to be outside the safe range, requiring the initiation of an adjustment process. When the pressure is determined to be outside the safe range, the onboard controller first calculates the pressure error value between the pressure reading and the median value of the safe range. The median value of the safe range is the arithmetic mean of the upper and lower limits. The pressure error value reflects the degree and direction of the current pressure deviating from the ideal operating point.

[0058] Subsequently, the airborne controller employs a proportional-integral-derivative (PID) control algorithm to calculate the adjustment amount of the actuator resistance based on the pressure error value. PID is an error feedback-based control method that calculates the required adjustment amount by integrating information from three aspects: the current error, historical error accumulation, and error change trend. First, the input to the PID controller is the pressure error value. The controller generates an output control quantity based on this error value, which is determined by three components: the first component is the proportional control component, which is proportional to the current pressure error value. When the pressure deviation is large, the proportional component increases accordingly, enabling the system to respond quickly to current pressure anomalies. The strength of the proportional component is determined by the proportional coefficient, which is preset according to the response characteristics of the hydraulic system. The second component is the integral control component, which comprehensively considers the cumulative effect of historical pressure errors. Even if the current error is small, if there has been a long-term historical deviation, the integral component will adjust accordingly. The integral component gradually accumulates and generates a continuous adjustment effect, used to eliminate steady-state pressure deviations caused by system leakage or load changes. The strength of the integral component is determined by the integral coefficient. The third component is the derivative control component, which makes predictive adjustments based on the changing trend of pressure error. When the pressure rises or falls rapidly, the derivative component increases the adjustment force in advance to suppress drastic pressure fluctuations and avoid overshoot or oscillation during the adjustment process. The strength of the derivative component is determined by the derivative coefficient. The above three components are superimposed inside the controller to form the final control output. After conversion, this output is used to adjust the resistance value of the actuator. The proportional, integral, and derivative coefficients need to be pre-tuned according to the dynamic characteristics of the hydraulic system and the specific operating scenario. For example, in the tillage operation scenario, the hydraulic system load changes drastically, so a larger proportional and derivative coefficient can be set to enhance the response speed and suppress fluctuations, while a moderate integral coefficient can be set to eliminate steady-state errors. In the sowing operation scenario, the hydraulic system load is relatively stable, so a smaller proportional and derivative coefficient can be set to avoid over-adjustment, while a smaller integral coefficient can be set to maintain system stability. By properly tuning these three coefficients, the proportional-integral-derivative controller can achieve rapid and stable control of hydraulic pressure under different operating conditions.

[0059] Subsequently, the airborne controller maps the control quantity output by the proportional-integral-derivative controller to a specific adjustment amount for the actuator resistance. It is proportional to the output control quantity q, that is ,in The resistance mapping coefficient is predetermined based on the actuator type and adjustment range. For actuators driven by hydraulic cylinders, the adjustment amount corresponds to the change in the hydraulic valve opening; for actuators driven by hydraulic motors, the adjustment amount corresponds to the change in the control current of the proportional pressure reducing valve. The onboard controller adjusts the adjustment amount accordingly. The actuator output resistance is fine-tuned in a closed loop. For example, when the pressure is too high, the actuator resistance is reduced by decreasing the hydraulic valve opening, thus reducing the load on the hydraulic system and lowering the pressure. When the pressure is too low, the actuator resistance is increased by increasing the hydraulic valve opening, thus increasing the load on the hydraulic system and raising the pressure. This adjustment process is continuously performed in discrete control cycles. In each control cycle, the onboard controller re-acquires the current pressure reading, calculates the new pressure error value, inputs it into the proportional-integral-derivative (PID) controller to calculate the new control quantity, and updates the actuator resistance adjustment accordingly. As the pressure error value gradually decreases, the output of the PID controller also decreases accordingly, forming a negative feedback closed loop. When the pressure reading gradually approaches the median of the safe range and eventually stabilizes within the preset safe range, for example, when the absolute value of the pressure error is less than 50 Pa, the onboard controller determines that the adjustment process is complete and stops active adjustment.

[0060] After adjustment, the onboard controller combines the current stable pressure reading, the adjusted actuator resistance value, and the convergence time of the adjustment process into stable parameters. These stable parameters reflect the optimal working state of the agricultural machine under the current operating conditions and are an important basis for subsequent path iteration optimization. Each agricultural machine's onboard controller uploads its own stable parameters to the central controller via a wireless communication module. The uploaded data includes the agricultural machine's identifier, timestamp, stabilized pressure reading, adjusted actuator resistance value, and the convergence time of the adjustment process.

[0061] By monitoring hydraulic pressure in real time and using a proportional-integral-derivative control algorithm to adjust actuator resistance in a closed loop, the system ensures that each agricultural machine can quickly return to a safe range when the load fluctuates, thus avoiding overload damage. At the same time, stable parameters are uploaded to the central controller to provide accurate equipment status data support for subsequent path iteration optimization.

[0062] Step S6: The central controller combines the historical operating data and stable parameters of each agricultural machine to evaluate the durability index of each agricultural machine. If any durability index is lower than the preset threshold, the collaborative path is updated by combining real-time multi-source data fusion to obtain the final multi-agricultural machine collaborative operation scheme.

[0063] Step S6 includes: the central controller obtains the cumulative working time and component wear estimate from the onboard controllers of each agricultural machine as historical operating data, and extracts the real-time operating condition information of each agricultural machine from stable parameters; the cumulative working time, component wear estimate and real-time operating condition information are normalized and weighted and fused to generate the durability index of each agricultural machine; it is determined whether the durability index of any agricultural machine is lower than the preset durability threshold. If so, the corresponding agricultural machine is marked as a high-risk agricultural machine for wear and tear, and real-time multi-source sensing data is obtained to update the vertical domain knowledge graph, iteratively optimize the collaborative path, and obtain the final multi-agricultural machine collaborative operation scheme.

[0064] Specifically, after the agricultural machinery convoy completes the closed-loop adjustment of hydraulic pressure and uploads the stable parameters to the central controller, the central controller integrates the historical operating data of each agricultural machine with the current stable parameters to conduct a comprehensive assessment of the durability status of the agricultural machinery convoy, and iteratively optimizes the collaborative path based on the assessment results, ultimately generating a final collaborative operation plan that adapts to the current status of each agricultural machine.

[0065] Specifically, the central controller first obtains historical operating data from the onboard controllers of each agricultural machine. This historical data includes cumulative working time and estimated component wear. The cumulative working time is recorded in the controller's operating log for each machine. At the end of each work cycle or in response to a query request from the central controller, the onboard controller uploads the cumulative working time to the central controller. This data reflects the total operating time of the agricultural machine since it was put into use and is a fundamental indicator for evaluating the overall lifespan of the machine. Each machine's key moving parts are equipped with vibration sensors that collect vibration acceleration signals in real time. The onboard controller performs a Fast Fourier Transform on the raw vibration signals to obtain a spectrum and extracts characteristic parameters such as total vibration energy and the proportion of high-frequency energy. It then compares these current characteristic parameters with a pre-stored standard wear curve. The standard wear curve records the vibration characteristics of the component at different wear levels. Pattern matching is used to find the point that best matches the current characteristics; the wear level corresponding to this point is the estimated component wear value. The onboard controller periodically uploads the estimated value to the central controller. The central controller also extracts real-time operating condition information related to the dynamic response characteristics of the hydraulic system from the stable parameters uploaded by each agricultural machine. For example, the convergence time of the adjustment process reflects the response sensitivity of the hydraulic system. The longer the convergence time, the more sluggish the system response, internal leakage or increased wear, resulting in a decrease in adjustment capability. Incorporating this time into the durability index calculation can identify potential failure risks of the hydraulic system earlier and improve the accuracy and comprehensiveness of durability assessment.

[0066] Furthermore, the central controller normalizes and weights the acquired cumulative working time, component wear estimates, and real-time operating condition information to generate a comprehensive durability index for each agricultural machine. Normalization eliminates the influence of differences in physical dimensions and magnitudes on the fusion result, mapping the cumulative working time, component wear estimates, and adjustment convergence time in the real-time operating condition information to the 0-1 range. After normalization, the central controller assigns weight coefficients to the three types of data. For example, the cumulative working time weight is set to 0.3, the component wear estimates weight is set to 0.5, and the real-time operating condition information weight is set to 0.2. This ensures that the component wear estimates, which best reflect the actual wear state of the mechanical structure, dominate the comprehensive index. The reciprocal of the weighted sum is used as the durability index for each agricultural machine. This index quantifies the current health status of each machine and its remaining health margin. A higher value indicates a better machine condition and longer remaining lifespan, while a lower value indicates the machine is closer to the end of its lifespan or has a higher risk of failure.

[0067] The central controller compares the durability index of each agricultural machine with a preset durability threshold. This threshold, pre-set based on the machine model, years of use, and maintenance history, is used to determine if a machine has entered a high-risk state. If a machine's durability index is higher than or equal to the preset threshold, it is considered to be in good condition and can continue normal operations. If the index is lower than the threshold, the machine is identified as entering a high-risk state, and the central controller marks it as such, triggering a collaborative path update process. When a high-risk machine is identified, the central controller immediately acquires real-time multi-source sensing data to update the vertical domain knowledge graph. This real-time multi-source sensing data includes newly collected soil hardness distribution data from a soil hardness sensor array, newly identified crop density distribution data from a high-resolution camera mounted on a drone, the latest real-time resistance distribution data obtained from the preceding steps, and the newly identified high-risk machine identifier. This data reflects the latest changes in the site environment and machine condition, forming the basis for iterative path optimization.

[0068] The central controller inputs the aforementioned real-time multi-source sensing data into the vertical domain knowledge graph for fusion and updating. The vertical domain knowledge graph stores plot grid nodes, soil hardness attributes, crop density attributes, agricultural machinery nodes, and various relationships in a graph structure. The update process includes: updating the hardness attribute of the corresponding plot grid node with newly collected soil hardness data; updating the density attribute of the corresponding plot grid node with newly identified crop density data; establishing a relationship between real-time resistance distribution data and the corresponding plot grid node; establishing a high-resistance relationship between the high-risk agricultural machinery identifier and its current location and the plot grid node it is about to visit, and appropriately increasing the weight of this relationship to indicate that these areas have a high operational risk for the agricultural machinery. After updating the vertical domain knowledge graph, the central controller iteratively optimizes the collaborative path based on the updated graph. The core of the path optimization is to recalculate the path cost function. In addition to considering soil hardness and crop density, a protection term for high-risk agricultural machinery is added. Specifically, for agricultural machinery marked as high-risk, when its path passes through a grid of plots with high resistance, the path cost will increase significantly. The increase is proportional to the weight of the relationship. This cost setting makes the path planning algorithm tend to avoid high-resistance areas for high-risk agricultural machinery.

[0069] Subsequently, the central controller employs the A* path search algorithm to replan the collaborative path for each machine in the formation from its current position to the boundary of the target work area. During the planning process, the algorithm uses the updated cost map as the search space, aiming to minimize path costs while satisfying constraints for multi-machine collaboration, including avoiding collisions between machines, maintaining the schedulability of the formation, and ensuring full coverage of the work area. For high-risk machines, the algorithm actively guides them to avoid high-resistance areas with higher costs; for machines with better durability, the algorithm allows them to undertake more work in high-resistance areas to balance the load of the entire formation. The replanned paths of each machine are integrated to form a new collaborative path scheme. This scheme includes the updated multi-machine trajectory sequence, the expected resistance distribution corresponding to each trajectory point, and differentiated workload allocation suggestions for machines with different durability conditions. The new scheme achieves proactive protection of high-risk machines while minimizing the impact on overall work progress, extending the overall fault-free operation time of the formation.

[0070] Finally, the central controller sends the final multi-machine collaborative operation scheme as an instruction set to the onboard controllers of each machine, guiding them to complete the remaining collaborative operation tasks. Each machine continues to operate according to the new path, and repeats the aforementioned attitude correction, load detection, formation adjustment and pressure regulation steps during the operation, forming a continuous closed-loop optimization.

[0071] In summary, this application provides a multi-machine collaborative operation method based on a vertical domain large model. By constructing a vertical domain knowledge graph and integrating soil hardness and crop density data, a high-precision load distribution prediction model is generated, avoiding the problem of uneven load at the source. This method collects multi-source data such as machine attitude, torque, and hydraulic pressure in real time, and uses intelligent algorithms such as proportional-integral-derivative control and genetic algorithms to dynamically optimize and adjust path following parameters, formation, and actuator resistance, realizing multi-level collaborative control from single-machine attitude correction to formation load balancing. Furthermore, a durability assessment mechanism is introduced, which integrates cumulative working time, component wear estimates, and hydraulic response characteristics to generate durability indicators, actively protecting high-risk machines and iteratively optimizing collaborative paths. Through hierarchical collaboration between the central controller and the onboard controller, an integrated intelligent collaborative system of perception, decision-making, and execution is constructed, significantly improving the operation accuracy, load balance, equipment safety, and formation stability of multiple machines in complex terrain, effectively extending the service life of the machines, and providing an innovative technical solution for the cluster collaborative operation of intelligent agricultural equipment.

[0072] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0073] If the integrated unit is implemented as 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 this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0074] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A method for multi-machine collaborative operation based on a large vertical domain model, characterized in that, The method includes: Step S1: The central controller receives multi-source sensing data uploaded by the airborne controller, integrates and constructs a vertical domain knowledge graph, obtains a load distribution prediction model, and generates an initial collaborative path to be sent to the airborne controllers of each agricultural machine. Step S2: The onboard controller of each agricultural machine collects its own attitude data in real time, calculates the attitude deviation from the initial collaborative path, and if it exceeds the preset threshold, it corrects the path following parameters and uploads them. The central controller then aggregates and generates an optimized collaborative path. Step S3: The central controller receives the torque values ​​collected by the load detection circuit of each agricultural machine, calculates the resultant torque distribution based on the relative positions of the agricultural machines in the optimized cooperative path, and determines the lever arm length distribution of each agricultural machine. Step S4: The central controller analyzes the center of gravity offset of the group based on the distribution of lever arm length and the weight of the agricultural machinery. If the group falls into the preset high load area, the central controller generates a formation change command and issues it. Each onboard controller responds to the command and readjusts the relative position of the machine. Step S5: Each onboard controller monitors the hydraulic pressure of the machine in real time. If the pressure reading of any agricultural machine exceeds the safe range, the actuator resistance is finely adjusted and the stable parameters are uploaded to the central controller. Step S6: The central controller combines the historical operating data and stable parameters of each agricultural machine to evaluate the durability index of each agricultural machine. If any durability index is lower than the preset threshold, the collaborative path is updated by combining real-time multi-source data fusion to obtain the final multi-agricultural machine collaborative operation scheme.

2. The multi-machine collaborative operation method based on a large vertical domain model according to claim 1, characterized in that, Step S1 involves fusing and constructing a vertical domain knowledge graph to obtain a load distribution prediction model, including: Soil hardness data and crop density image data from multiple points within the plot are collected as multi-source sensing data. Spatiotemporal alignment and feature-level fusion are performed on the multi-source sensing data to obtain a fused feature vector. Based on the fused feature vector, a vertical domain knowledge graph is constructed, which includes plot spatial grid nodes, soil hardness attributes, crop density attributes, and their interrelationships. Based on the node attributes and edge relationships in the vertical domain knowledge graph, the load influence factors of each grid area are deduced. Based on the load influence factor, a load distribution prediction model incorporating spatial heterogeneity characteristics is obtained.

3. The multi-machine collaborative operation method based on a large vertical domain model according to claim 2, characterized in that, Step S1 generates the initial cooperative path, including: The regions with soil hardness exceeding a preset hardness threshold and regions with crop density exceeding a preset density threshold are extracted from the load distribution prediction model as comprehensive restricted areas. The target work area is divided into grids, and the comprehensive restricted area is marked as an insurmountable obstacle; A path planning algorithm is used, combined with the vertical domain knowledge graph, to generate an initial collaborative path that bypasses the comprehensive restricted area and covers the target work area.

4. The multi-machine collaborative operation method based on a large vertical domain model according to claim 1, characterized in that, Step S2 includes: The onboard controller of each agricultural machine collects the attitude data of each agricultural machine in real time through the attitude sensor installed on the machine. Calculate the deviation vector between the actual posture of each agricultural machine and the given posture in the initial cooperative path; The magnitude of the deviation vector is calculated and it is determined whether the magnitude exceeds the preset attitude deviation threshold. If it does, the proportional-integral-derivative control algorithm is used to adjust the path following parameters of the machine to obtain the corrected path following parameters. The onboard controllers of each agricultural machine upload the corrected path following parameters to the central controller, which then aggregates and generates an optimized collaborative path based on the current location of each agricultural machine and the corrected parameters.

5. The multi-machine collaborative operation method based on a large vertical domain model according to claim 1, characterized in that, Step S3 includes: Each agricultural machine, through a load detection circuit, collects the torque value of its working parts in real time via a torque sensor installed on the tillage device, and uploads the torque value to the central controller. The central controller extracts the relative position coordinates of each agricultural machine in the formation from the optimized collaborative path and constructs a position relationship matrix that includes lateral spacing and longitudinal offset. Using the torque value and the position relationship matrix as input, a genetic algorithm is used to iteratively optimize and calculate the resultant torque distribution formed after the superposition of multiple machine loads. The lever arm length distribution of each agricultural machine in the formation is determined based on the resultant moment distribution.

6. The multi-machine collaborative operation method based on a large vertical domain model according to claim 1, characterized in that, Step S4 includes: The central controller calculates the overall center of gravity position of the multi-machine group based on the lever arm length distribution and the mass parameters of each agricultural machine; Compare the overall center of gravity position with the initial reference center of gravity position to calculate the group center of gravity offset; Determine whether the center of gravity offset of the group falls into the preset high load area. If it does, the central controller generates a formation change instruction that includes lateral spacing adjustment and longitudinal misalignment adjustment, and sends it to the onboard controller of each agricultural machine. The onboard controllers of each agricultural machine respond to the commands, perform adaptive formation transformation, adjust the lateral spacing and longitudinal misalignment of the machine in the multi-machine formation, and redistribute the relative posture.

7. The multi-machine collaborative operation method based on a large vertical domain model according to claim 1, characterized in that, Step S5 includes: S51: The onboard controller of each agricultural machine collects the real-time pressure readings of the machine's hydraulic system at a preset sampling frequency through pressure sensors installed on the hydraulic pipeline, forming a pressure reading sequence; S52: Compare the current pressure reading in the pressure reading sequence with the preset safe pressure upper limit and the preset safe pressure lower limit to determine whether it exceeds the safe range. If it does, calculate the pressure error value between the pressure reading and the median value of the safe range. S53: Using a proportional-integral-derivative control algorithm, the adjustment amount of the actuator resistance is calculated based on the pressure error value, and the output resistance of the corresponding actuator of the machine is finely adjusted in a closed loop. S54: Repeat steps S51 to S53 until the current pressure reading stabilizes within the preset safety range to obtain stable parameters; S55: The onboard controller of each agricultural machine uploads the stability parameters to the central controller.

8. The multi-machine collaborative operation method based on a large vertical domain model according to claim 1, characterized in that, Step S6 includes: The central controller obtains the cumulative working time and component wear estimate from the onboard controllers of each agricultural machine as historical operating data, and extracts the real-time operating condition information of each agricultural machine from the stable parameters. The cumulative working time, the estimated wear of the components, and the real-time working condition information are normalized and weighted and fused to generate durability indicators for each agricultural machine. If any agricultural machine's durability index is lower than a preset durability threshold, the corresponding agricultural machine is marked as a high-risk agricultural machine for wear and tear. Real-time multi-source sensing data is obtained, the vertical domain knowledge graph is updated, the collaborative path is iteratively optimized, and the final multi-agricultural machine collaborative operation scheme is obtained.

Images