Remote linkage and automatic control optimization method and system for agricultural equipment

Abstract

The application discloses an agricultural equipment remote linkage and automatic control optimization method and system, and belongs to the technical field of automatic control of agricultural equipment. The method comprises the following steps: collecting real-time sensing data streams of multiple agricultural equipment in a working area, which comprises equipment operation state parameters, environmental factor monitoring values and soil physicochemical index sequences; performing spatio-temporal correlation aggregation processing on the real-time sensing data streams to generate a coupling degree matrix between the equipment, the matrix comprising mutual influence coefficients and delay response times between the equipment; constructing a multi-equipment collaborative control constraint model based on the coupling degree matrix, which comprises equipment operation priority, energy consumption limitation and working area coverage requirement; calling an optimization solver to perform iterative search processing on the model to generate an optimal linkage control instruction sequence; distributing the instruction sequence to corresponding equipment through a remote communication network for execution, and collecting feedback data in real time to update the coupling degree matrix.

Description

Technical Field

[0001] This invention relates to the field of automatic control technology for agricultural equipment, specifically to a method and system for optimizing remote linkage and automatic control of agricultural equipment. Background Technology

[0002] In large-scale agricultural production, various automated agricultural machines, such as irrigation equipment, fertilization equipment, tillage equipment, and plant protection equipment, are typically deployed within the work area. Each piece of equipment operates independently, with its operating parameters set based on data collected by its own sensors. There is a lack of effective information exchange and coordination mechanisms between the devices. Existing technologies sometimes use preset timing logic to achieve simple sequential start-stop linkage between devices, such as delaying the start of fertilization equipment after irrigation. However, this control method does not consider the real-time disturbances to soil, microclimate, and other environmental factors caused by equipment operation, and their cross-influence on the operating effects of other equipment. When equipment operates close to each other, if one piece of equipment changes soil moisture or local temperature and humidity, the preset operating parameters of adjacent equipment become inapplicable to the changed site conditions. Each piece of equipment still executes according to its original plan, leading to uneven work quality, resource waste, and even operational conflicts. The deficiency in existing technologies lies in the fact that the correlations relied upon for multi-device collaborative control are mostly static settings or based on simple rule derivations, failing to reflect the dynamic evolution of coupling strength and response delays between devices due to environmental changes. Meanwhile, the generation of control commands often relies on optimization calculations at a single point in time, lacking a closed-loop mechanism to continuously correct the coupling relationship between devices using feedback data after execution. When the operating environment changes significantly due to factors such as sudden weather changes or soil heterogeneity, the original control model cannot adaptively adjust, and the optimization results deviate from the actual operating requirements, making it difficult to maintain the optimal balance between energy efficiency and completion time for multi-device collaborative operation under dynamic conditions.

[0003] The technical problem to be solved by this invention is how to accurately characterize the intensity of mutual influence and time lag between multiple devices in a dynamically changing farmland environment, and how to use the measured data continuously transmitted back during the operation to iteratively update the coupling relationship between the devices, so that the linkage control commands can continuously track environmental changes. Summary of the Invention

[0004] The purpose of this invention is to provide a method and system for optimizing remote linkage and automatic control of agricultural equipment, so as to solve the problem that the coupling relationship of multiple devices working together is static and fixed and lacks the ability to correct the closed loop after execution in the prior art, and realize the continuous optimization of linkage control between devices in dynamic environment.

[0005] To achieve the above objectives, the present invention provides the following technical solution: The present invention relates to a method and system for optimizing remote linkage and automatic control of agricultural equipment, and provides the following technical solution to solve the problem of difficulty in balancing dynamic environmental changes and optimal operation efficiency during multi-device collaborative operation.

[0006] This invention provides a method for optimizing remote linkage and automatic control of agricultural equipment, comprising: collecting real-time sensor data streams from multiple agricultural devices within a work area, wherein the real-time sensor data streams include equipment operating status parameters, environmental factor monitoring values, and soil physicochemical index sequences; performing spatiotemporal correlation aggregation processing on the real-time sensor data streams to generate a coupling degree matrix between the devices, wherein the coupling degree matrix includes mutual influence coefficients and delay response times between the devices; constructing a multi-device collaborative control constraint model based on the coupling degree matrix, wherein the multi-device collaborative control constraint model includes equipment operation priorities, energy consumption limits, and work area coverage requirements; calling an optimization solver to perform iterative search processing on the multi-device collaborative control constraint model to generate an optimal linkage control command sequence; distributing the optimal linkage control command sequence to the corresponding devices for execution via a remote communication network, and collecting feedback data in real time to update the coupling degree matrix. Through the above process, the spatiotemporal coupling relationship between devices is quantified into a matrix, and a closed loop of optimization solution is dynamically introduced, enabling the linkage commands to adapt to real-time changes in the environment and equipment status, simultaneously improving overall work efficiency and reducing energy consumption while ensuring work coverage requirements.

[0007] As a further aspect of this invention, when generating the coupling degree matrix, the real-time sensing data stream is divided into multiple subsequences according to time windows, with each subsequence corresponding to a sampling period. For each subsequence, the operating status parameters and environmental factor monitoring values ​​of each device are extracted, the Pearson correlation coefficient between each pair of devices is calculated, and a spatial distance weight matrix is ​​calculated based on the geographical coordinates of the devices. The Pearson correlation coefficient and the spatial distance weight matrix are weighted and superimposed to generate the inter-device correlation strength matrix under each time window. The correlation strength matrices of multiple consecutive time windows are exponentially smoothed and fused to obtain the coupling degree matrix, and the delay response time is determined by the peak position of the cross-correlation function. This processing method effectively integrates the operating status correlation in the time dimension and the geographical distribution influence in the spatial dimension, making the linkage coupling relationship between devices more accurate and providing a refined quantitative basis for subsequent collaborative decision-making.

[0008] As a preferred embodiment of the present invention, when constructing a multi-device collaborative control constraint model, device pairs with mutual influence coefficients greater than a preset threshold are extracted from the coupling degree matrix to construct a strongly correlated device sub-network graph, which includes node attributes and edge weights. A collaborative controller is assigned to each strongly correlated device sub-network graph, and the collaborative controller maintains the joint state vector of all devices within the sub-network. The optimization weights of each sub-network are assigned according to the device operation priority. Combining energy consumption constraints and operation area coverage requirements, a multi-objective optimization function is established with the objectives of minimizing operation completion time and optimizing total energy consumption. The multi-objective optimization function is combined with the delay response time in the coupling degree matrix to form a multi-device collaborative control constraint model. This model decouples a large-scale device group into strongly correlated sub-networks for distributed management and control, while incorporating delay response into the constraints, making the generated collaborative control scheme more consistent with the temporal coupling characteristics in actual operations, which is beneficial to improving the real-time performance and coordination of linkage control.

[0009] As a specific implementation of this invention, the process of calling the optimization solver to generate the optimal linkage control command sequence includes: initializing the population of individuals in the optimization solver, each individual representing a set of control parameter vectors, wherein the control parameter vectors include the start-up timing, operating power, and movement path deflection angle of each device; substituting each individual into the multi-device collaborative control constraint model, calculating the fitness value of the individual, wherein the fitness value is the weighted sum of the multi-objective optimization function and the superposition of the constraint violation penalty term; performing selection, crossover, and mutation operations on the population based on the fitness value to generate a new generation of population and recording the current best individual; repeating the iteration until the convergence condition is met, and decoding the current best individual into the optimal linkage control command sequence. By combining evolutionary search with constraint penalty mechanism, it is possible to efficiently find the optimal solution in a complex parameter space and obtain linkage control commands that take into account operational efficiency, energy consumption, and constraint compliance.

[0010] Preferably, during population initialization, several sets of control parameters are randomly selected from a historical operation database as initial parents. This historical operation database contains records of control parameters from past successful operations. The initial parents are then subjected to Gaussian perturbation to generate offspring individuals of equal number to the original parents. The variance of the Gaussian perturbation is positively correlated with the coefficient of variation of the current environmental factor monitoring values. The initial parents and offspring individuals are then merged to form the initial population. Utilizing historical effective experience to guide the search starting point while adaptively adjusting the perturbation intensity according to the magnitude of environmental fluctuations helps enhance population diversity during periods of drastic environmental change and accelerates convergence to a control strategy adapted to new operating conditions.

[0011] As a technical solution of this invention, the instruction distribution and feedback update process is as follows: the optimal linkage control instruction sequence is split into single-device control instructions according to the device identifier, and sent to the edge gateway of each device via the MQTT protocol; after parsing the instructions, the edge gateway of each device drives the actuator to act, and at the same time collects the actual operating parameters and environmental factor monitoring values ​​after execution at a fixed frequency as feedback data; the feedback data is uploaded to the cloud server, the cloud server replaces the data in the corresponding time window in the real-time sensor data stream with the feedback data, re-executes the spatiotemporal correlation aggregation processing, and generates an updated coupling degree matrix. Based on a lightweight IoT protocol, reliable instruction distribution and execution status feedback are achieved, and the coupling degree matrix is ​​continuously corrected by the measured feedback data, enabling the system to have adaptive capabilities to changes in the actual response characteristics of the devices.

[0012] As a further improvement of this invention, after real-time acquisition of feedback data, a deviation vector between the feedback data and the expected parameters in the optimal linkage control command sequence is calculated. This deviation vector contains the difference between the actual and target values ​​of each parameter. The deviation vector is then input to an online learning module, which uses an incremental random forest algorithm to update the parameters in the multi-device collaborative control constraint model. These parameters include equipment operation priority weights and energy consumption limit thresholds. By identifying the deviation between the expected and actual values ​​and adjusting the model parameters online, the control model can automatically calibrate to keep pace with the slow evolution of the operating environment and equipment performance, maintaining long-term stability of the optimized control effect.

[0013] Preferably, in the optimization process, the crossover operation uses a simulated binary crossover operator, and the mutation operation uses a polynomial mutation operator, with the mutation probability decreasing linearly with the number of iterations. This combination of operators and mutation strategy maintains strong global exploration capabilities in the early stages of the search and enhances local exploration accuracy in the later stages, facilitating a faster approximation of the globally optimal linkage control command.

[0014] Furthermore, in the incremental random forest algorithm, each decision tree is incrementally trained only on parameter samples whose absolute value in the deviation vector exceeds a preset threshold during updates, and the maximum depth of the tree is dynamically adjusted. Learning only from significantly biased samples reduces unnecessary computation, avoids frequent fluctuations in model parameters due to minor noise, and enhances sensitivity to changes in key operating conditions while ensuring the timeliness of model updates.

[0015] This invention also provides a remote linkage and automatic control optimization system for agricultural equipment, including a processor and a memory connected to the processor. The memory stores programs, instructions, or code, and the processor executes the programs, instructions, or code in the memory to implement the aforementioned remote linkage and automatic control optimization method for agricultural equipment. This system integrates sensor data aggregation, model building, optimization solution, instruction distribution, and online learning into a closed-loop control framework. It can autonomously generate and execute optimal linkage strategies that balance efficiency and energy consumption for heterogeneous agricultural machinery in multi-regional collaborative operation scenarios, and continuously evolve the model based on operational feedback, effectively improving the intelligence level and resource utilization efficiency of large-scale precision agriculture operations.

[0016] The technical effects and advantages provided by the present invention in the above technical solution are as follows: The real-time sensor data stream is divided into subsequences by time windows. Operating status parameters and environmental factor monitoring values ​​of each device are extracted, and Pearson correlation coefficients between each pair of devices are calculated. Simultaneously, a spatial distance weight matrix is ​​calculated based on the geographical coordinates of the devices. The Pearson correlation coefficients and the spatial distance weight matrix are weighted and superimposed to generate a device correlation strength matrix for each time window. Then, the correlation strength matrices of multiple consecutive time windows are exponentially smoothed and fused to obtain a coupling degree matrix, where the delay response time is determined by the peak position of the cross-correlation function. This processing method integrates the strength of mutual influence between devices with the delay propagation characteristics in the spatiotemporal dimensions into the coupling degree matrix. This allows the matrix to not only reflect whether there is a correlation between devices but also quantify the attenuation of correlation at different spatial distances and the time lag required for influence transmission. Compared to conventional techniques that rely solely on a single correlation coefficient or manual experience to define coupling relationships, the resulting coupling degree matrix enables the multi-device collaborative control constraint model to dynamically adjust the constraint strength between devices based on their actual locations and local environments. The optimal linkage control command sequence searched by the optimizer on this basis more closely matches the actual coupling state of each area on-site, avoiding overshooting or response lag caused by neglecting spatial attenuation and time lag, thus improving the execution matching degree of linkage commands in dynamic farmland environments. After the optimal linkage control command sequence is distributed to the devices for execution via a remote communication network, actual operating parameters and environmental factor monitoring values ​​are collected at a fixed frequency as feedback data. This feedback data is uploaded to a cloud server, replacing the data in the corresponding time window of the real-time sensor data stream, and spatiotemporal correlation aggregation processing is re-executed to generate an updated coupling degree matrix. Furthermore, the deviation vector between the feedback data and the expected parameters in the command sequence is calculated. This deviation vector is input into an online learning module, and an incremental random forest algorithm is used to update the device operation priority weights and energy consumption limit thresholds in the multi-device collaborative control constraint model. This technique directly incorporates the actual results of the operation into the synchronous update loop of the coupling relationship and constraint model. The feedback data is used both to correct the coupling degree matrix between devices and to dynamically adjust the control parameters within the constraint model. Compared to methods that are optimized once and then not corrected, or rely solely on manual parameter tuning based on operator experience, this mechanism combining closed-loop feedback and online learning enables the linkage control model to autonomously perceive operational deviations and respond instantly. It continuously tracks system characteristic drift caused by gradual changes in soil conditions, equipment wear, or external climate disturbances. The multi-device collaborative control constraint model always maintains a state that reflects the current actual physical process. Subsequent control cycle command optimization is based on the latest coupling relationships and control parameters, fundamentally suppressing the accumulation of optimization deviations caused by model mismatch. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0018] Figure 1 This is a flowchart of the optimization method for remote linkage and automatic control of agricultural equipment; Figure 2 This is a flowchart for calculating the coupling degree matrix and delay response time of agricultural equipment based on spatiotemporal fusion. Figure 3 This is a flowchart of the construction process of a multi-device collaborative control constraint model based on the coupling degree matrix; Figure 4 This is a flowchart for optimizing the control parameters of multiple agricultural devices in a coordinated manner. Figure 5 This is a flowchart of the distribution of collaborative control commands and online model updates for agricultural equipment; Figure 6 This is a flowchart simulating binary crossover and polynomial mutation operations. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] See Figure 1 This invention provides a method for optimizing remote linkage and automatic control of agricultural equipment, comprising: collecting real-time sensor data streams from multiple agricultural devices within a work area, wherein the real-time sensor data streams include equipment operating status parameters, environmental factor monitoring values, and soil physicochemical index sequences; performing spatiotemporal correlation aggregation processing on the real-time sensor data streams to generate a coupling degree matrix between the devices, wherein the coupling degree matrix includes mutual influence coefficients and delay response times between the devices; constructing a multi-device collaborative control constraint model based on the coupling degree matrix, wherein the multi-device collaborative control constraint model includes equipment operation priority, energy consumption limits, and work area coverage requirements; calling an optimization solver to perform iterative search processing on the multi-device collaborative control constraint model to generate an optimal linkage control command sequence; distributing the optimal linkage control command sequence to the corresponding devices for execution via a remote communication network, and collecting feedback data in real time to update the coupling degree matrix.

[0021] Real-time sensor data streams are collected sequentially by various sensors deployed on agricultural equipment such as tractors, harvesters, and irrigation valve assemblies. Equipment operating status parameters may include engine speed, operating torque, and pesticide flow rate; environmental factor monitoring values ​​may include air temperature and humidity, and light intensity; and soil physicochemical index sequences reflect changes in soil moisture content, electrical conductivity, and nitrogen, phosphorus, and potassium content over time. The collected data is aggregated to edge computing nodes or the cloud, forming a continuously updated data stream.

[0022] In the spatiotemporal correlation aggregation process, the dynamic mutual interference between operational behaviors is quantified by extracting the temporal correlation and spatial proximity relationships between devices. The resulting coupling degree matrix not only reflects the intensity of the influence between devices at the current moment but also calibrates the delay response time, which is used to characterize the time delay for a command to be transmitted from one device to another to take effect. This matrix provides a basis for the correlation topology and dynamic characteristics for collaborative modeling.

[0023] The multi-device collaborative control constraint model is established based on the strongly correlated structure revealed by the coupling degree matrix. It decomposes the complex multi-machine system into several cooperative subnets. Under the constraints of job scheduling, energy consumption limits, and full regional coverage, a multi-objective optimization problem is constructed that considers both completion time and total energy consumption. An optimization solver swarm intelligently searches for the optimal solution to this model, outputting linkage commands that include the startup sequence, power output, and path deflection angle sequence of each device. These commands are sent to the device side for execution via a remote network. The feedback data generated during execution is fed back, triggering dynamic adjustments to the coupling degree matrix, forming a closed-loop adjustment. Example 1:

[0024] In specific implementation, please refer to Figure 2 The duration of the time window is set based on the operating cycle of agricultural equipment and the frequency of changes in environmental factors. One approach is to set the time window length as an integer multiple of the equipment sensor sampling cycle, with the integer multiple being no less than 60. This ensures that a single time window contains at least 60 sampling points, guaranteeing the statistical reliability of the Pearson correlation coefficient calculation. The real-time sensor data stream is divided into multiple subsequences according to the same time window length, with each subsequence corresponding one-to-one with a continuous sampling period.

[0025] For a subsequence, extract the equipment operating status parameters and environmental factor monitoring values ​​for each agricultural device within the subsequence. Taking engine speed as an example among the equipment operating status parameters, the equipment... Within the subsequence The instantaneous engine speed value at each sampling point is denoted as ,in, For device identification index, The sampling point number, Increasing from 1 to , This represents the total number of sampling points within the subsequence. Correspondingly, the device can be represented in vector form. The engine speed sequence within this subsequence. The same extraction method is applied to environmental factor monitoring values, such as air temperature and soil moisture content; the sequences of the corresponding parameters for each device within the subsequence are extracted.

[0026] For any two devices within a subsequence, for example, device With equipment The Pearson correlation coefficient is used to quantify the linear correlation between parameter sequences of two devices. The calculation of the Pearson correlation coefficient involves the ratio of the covariance of each parameter to its respective standard deviation. First, the equipment... The mean of the parameter sequence within the subsequence and equipment The mean of the parameter sequence within the subsequence The parameter sequence can be a fusion sequence of one or more of the equipment operating status parameters and environmental factor monitoring values. The fusion method can be the first principal component sequence obtained by principal component analysis. Then, the equipment... Parameter sequence and device The parameter sequence at each sampling point The sum of the products of the deviations is the numerator; the denominator is the equipment. Standard deviation of parameter series and equipment The product of the standard deviations of the parameter series. The Pearson correlation coefficient has a range of [-1, 1], where close to 1 represents a positive correlation, close to -1 represents a negative correlation, and close to 0 represents no linear correlation.

[0027] A spatial distance weight matrix is ​​constructed synchronously based on the geographical coordinates of each device. With equipment The geographic coordinates are denoted as follows: and Euclidean distance between the two devices The distance is calculated after converting the latitude and longitude difference to planar distance. The equipment is included in the spatial distance weight matrix. With equipment Corresponding elements Determined via a negative exponential function:

[0028] in, Indicates the device within the current time window. With equipment Elements of the correlation strength matrix between them; Indicates device With equipment Pearson correlation coefficient within subsequences; Indicates device With equipment The corresponding spatial distance weight matrix elements; This represents the fusion coefficient, with a value range of [value range missing]. Fusion coefficient The value is determined through offline parameter optimization. One method is to select historical sensor data from multiple consecutive operating cycles as a sample set and set alternative parameters. Values ​​range from 0.1 to 0.9 with a step size of 0.1, each using an alternative... The corresponding correlation strength matrix is ​​generated to predict the influence strength between devices. The matching error between the predicted value and the actual observed device status linkage delay time is used as an indicator to select the candidate that minimizes the matching error. Value. A value determined in the above manner. The value is 0.7, and its selection is based on the following: in the sample set, when When the value is 0.7, the strong correlation equipment pairs predicted by the correlation strength matrix have the highest matching rate with equipment pairs with linkage delay of less than 3 seconds in actual operation, and the mean absolute error is the smallest.

[0029] Spatial distance weight elements By negative exponential function The calculation yielded that, is the distance normalization constant. The value can be set to the median of the distance between all devices, or according to the communication coverage radius of the devices. In one implementation, 50% of the effective operating distance of the communication modules between devices is used as the threshold. This allows devices within the communication range to receive a higher spatial weight, while the spatial weight of devices outside the range rapidly diminishes.

[0030] The above steps are applied to pairwise combinations of all devices within the subsequence to generate the device association strength matrix for the current time window. The matrix dimension equals the total number of devices, and the element at position (p, q) in the matrix is ​​the [device association strength matrix]. .

[0031] Exponential smoothing is applied to the correlation strength matrices of multiple consecutive time windows to obtain the coupling degree matrix. Let the k-th... The correlation strength matrix for each time window is as follows The fused coupling matrix Updated recursively. During the recursion, the new coupling degree matrix is ​​obtained by a weighted sum of the current time window association strength matrix and the previous coupling degree matrix, with a smoothing constant. Determine the extent to which historical data is retained. The value ranges from 0 to 1. The closer the value is to 1, the greater the weight of the current data. The closer to 0, the greater the weight of the historical smoothing value. One method for determining this is to adaptively adjust based on the degree of change in equipment status: when the rate of change of the Frobenius norm of the correlation strength matrix exceeds 30% over multiple consecutive time windows, Use 0.9 for rapid response to mutations; when the rate of change is below 10%, A value of 0.5 is used to enhance the smoothing effect. After recursively fusing multiple time windows, each element in the final coupling matrix represents the mutual influence coefficient between the corresponding devices.

[0032] The delayed response time is obtained through cross-correlation function analysis. For the device... parameter sequence With equipment parameter sequence Cross-correlation function With a delay step count of The calculation method for time is as follows: ,when A positive value indicates the device Parameters lag behind equipment parameter Each sampling point, when A negative value indicates the device Parameters are ahead. This applies to all calculated delay steps. In the process, find the value of the cross-correlation function that maximizes its absolute peak value. Value, denoted as .Will Multiply by the sampling interval to obtain the device. With equipment Response time between The unit is seconds. The latency response time is stored as an attribute in the corresponding position of the coupling matrix, and together with the mutual influence coefficient, it constitutes complete device coupling information.

[0033] Example 2: In specific implementation, please refer to Figure 3 Extract all mutual influence coefficients from the generated coupling matrix. Then, compare each mutual influence coefficient with a preset threshold. Compare. Preset threshold. The value is set to 0.6, based on the following: when the mutual influence coefficient between two agricultural devices is less than 0.6, the operational linkage strength of the two agricultural devices is considered insufficient to be included in mandatory collaborative control, and they are regarded as a loosely coupled relationship; when the mutual influence coefficient is greater than 0.6, it indicates that there is a strong correlation between the two agricultural devices, and synchronization constraints need to be maintained in collaborative control. All device pairs that satisfy the condition that the mutual influence coefficient is greater than the preset threshold ε are extracted. Each agricultural device participating in the device pair is taken as a node, and the device pair is taken as a connecting edge to construct a strongly correlated device sub-network graph. In the strongly correlated device sub-network graph, each node stores node attributes, including device type identifier, device rated power value, and device current state snapshot. The device current state snapshot is taken from the latest sampled value of the device's operating state parameters. Each edge in the strongly correlated device sub-network graph stores an edge weight, which consists of two values: the mutual influence coefficient between the two agricultural devices and the corresponding delay response time between the two agricultural devices.

[0034] For each strongly correlated device sub-network graph constructed, an independent cooperative controller is set up. Each cooperative controller internally maintains a joint state vector, which covers all agricultural devices within its strongly correlated device sub-network graph. The joint state vector records the device operating status, device location coordinates, and device movement trajectory of each agricultural device within the strongly correlated device sub-network graph in a time-series dimension. The device operating status includes two fields: engine speed and operating torque. The device location coordinates consist of longitude and latitude values ​​output by the Global Positioning System module at fixed time intervals. The device movement trajectory consists of a continuously set number of device location coordinate sequences, with the sequence length set to the coordinate values ​​of the most recent 60 sampling points.

[0035] Optimization weights are assigned to each strongly correlated equipment sub-network graph based on its operational priority. The operational priority is determined by two factors: the first is the crop type corresponding to the task, which is categorized into high-value crops and ordinary crops according to a pre-defined value level table; the second is the remaining duration of the operational period. Strongly correlated equipment sub-network graphs with a remaining operational period of less than 24 hours are assigned the highest optimization weight, while those with high-value crops are assigned the second-highest optimization weight. The remaining strongly correlated equipment sub-network graphs are linearly mapped to a range of 0.1 to 0.5 based on the reciprocal of their remaining operational period, thus obtaining the optimization weight for each strongly correlated equipment sub-network graph.

[0036] Construct a multi-objective optimization function that combines the two optimization objectives of task completion time and total energy consumption with weights. The specific form is as follows:

[0037] in, This represents the total value of the weighted objective function to be minimized; Indicates the index number of the strongly associated device sub-network diagram; This represents the sub-network diagram assigned to strongly related devices. Optimize the weights; A trade-off factor representing the target completion time for the task; This represents the compromise factor for the total energy consumption target, and and satisfy ; Represents a subnetwork diagram of strongly related devices. The total time taken for all agricultural equipment to complete all designated tasks, expressed in seconds; Represents a subnetwork diagram of strongly related devices. Total energy consumption of all agricultural equipment performing all operational tasks, expressed in kilowatt-hours. Compromise factor. and The value is dynamically set based on the current energy price index: when the energy price index exceeds 1.2 times the average agricultural electricity price over the past year, Take 0.3, 0.7; when the energy price index does not exceed 1.2 times the average agricultural electricity price over the past year, Take 0.7, Take 0.3.

[0038] Constraints are introduced during the solution of the multi-objective optimization function. These constraints include power constraints and coverage constraints. The power constraint stipulates that the operating power of each agricultural device within the strongly correlated sub-network graph at any given time does not exceed the rated power limit corresponding to that device. The coverage constraint stipulates that the union of the areas of the operation coverage polygons of all agricultural devices within the strongly correlated sub-network graph must account for at least 98% of the area of ​​the target plot polygon, which is defined by the coordinate sequence of the boundary vertices of the operation area.

[0039] The multi-objective optimization function is combined with the delay response time in the coupling matrix. The method of combining the two is as follows: in the time-series constraint set of the multi-device cooperative control constraint model, for each pair of devices whose delay response time in the coupling matrix is ​​not zero... and equipment Add a delay compensation term. The delay compensation term constrains the equipment. State adjustment time satisfy ,in For equipment The moment when a control command is issued and the state changes. The devices recorded in the coupling matrix With equipment The delay response time between them. Power constraints, coverage constraints, and timing constraints including delay compensation terms are incorporated into the constraint equations of the multi-objective optimization function to form a complete multi-device cooperative control constraint model. Example 3:

[0040] In specific implementation, please refer to Figure 4 The optimization solver maintains a population, where each individual corresponds to a set of control parameter vectors. Each control parameter vector is composed of three encoded parts: the first part is the startup timing encoding segment for each agricultural device, using real-number encoding, where each real number represents the startup delay in seconds for the corresponding agricultural device from time zero; the second part is the operating power encoding segment for each agricultural device, also using real-number encoding, where each real number represents the constant operating power maintained by the corresponding agricultural device during operation, in kilowatts; and the third part is the deflection angle encoding segment for each agricultural device's movement path, also using real-number encoding, where each real number represents the deflection angle of the corresponding agricultural device relative to a preset reference heading, in radians.

[0041] During population initialization, the optimization solver accesses the historical job database. This database stores control parameter records from past successful jobs. Each record contains historical values ​​for the start-up sequence, operating power, and movement path deflection angle of each agricultural device under a specific job task. The optimization solver randomly selects several control parameter records from the database, the number equal to the preset initial parent generation size. This initial parent generation size is set to 50. This setting is based on a trade-off between increased computational overhead due to an excessively large population size and insufficient search diversity due to an excessively small population size. Using 50 as the initial parent generation size allows the optimization solver to balance the large-scale search requirements in agricultural equipment collaborative control scenarios with the computing power of edge computing node processors. Each extracted control parameter record directly serves as an initial parent individual, forming the initial parent population.

[0042] For each initial parent individual in the initial parent population, a Gaussian perturbation is applied to generate offspring individuals of the same number as the original parents. The specific steps of the Gaussian perturbation are as follows: For each encoded segment of the control parameter vector of the initial parent individual, a Gaussian noise vector with the same dimension as the encoded segment is generated sequentially. Each element in the Gaussian noise vector independently follows a normal distribution with a mean of zero and a variance of σ². The Gaussian noise vector is then added element-wise to the corresponding encoded segment of the initial parent individual to obtain the value of the encoded segment of the offspring individual. The variance of the Gaussian perturbation is... The coefficient of variation is determined in real time based on the current environmental factor monitoring values. Variance is defined as the ratio of the standard deviation to the mean of the current environmental factor monitoring value series. The calculation method is as follows ,in, As a scaling factor, The value is set to 0.05, based on the principle of controlling the amplitude of the Gaussian perturbation to 5% of the intensity of environmental factor fluctuations. This ensures that offspring individuals inherit the superior control patterns of their parents while introducing reasonable variations that match the current environmental uncertainties. The generated offspring individuals are then merged with the initial parent individuals to form an initial population of 100 individuals.

[0043] Each individual in the initial population is sequentially substituted into the multi-device cooperative control constraint model to calculate the fitness value of each individual. The specific form of the fitness function is:

[0044] in, This represents the fitness value of an individual. This represents the weighted total value of the multi-objective optimization function calculated after substituting the current individual control parameter vector. The multi-objective optimization function includes the combined value of the operation completion time and total energy consumption under the weighted optimization weights of each strongly correlated sub-network graph. Indicates the penalty factor; Indicates the index number of the constraint condition; This represents the violation function of the l-th constraint under the current individual control parameter vector. The constraint violation function is defined as the difference between the actual value of the constraint and the constraint limit value. When the actual value satisfies the constraint... Less than or equal to zero, when the actual value exceeds the constraint limit. Greater than zero; This represents the larger value between zero and the constraint violation function value; that is, the penalty term is calculated only for constraint violations. Penalty Factor The value of is determined as follows: the maximum value of the constraint violation function of all individuals in the initial population is denoted as . The multi-objective optimization function of all individuals in the initial population The standard deviation of the value is denoted as Punishment factor The calculation method is as follows ,when When it is zero Set to zero.

[0045] Based on the calculated fitness values ​​of all individuals, selection, crossover, and mutation operations are performed on the population to generate a new generation. The selection operation employs a tournament selection strategy: a preset number of individuals are randomly selected from the current population to form a tournament candidate group (each group containing 3 individuals). The individual with the lowest fitness value from the tournament candidate group is selected to enter the next generation. This tournament selection operation is repeated until the number of individuals selected in the next generation reaches the population size. The crossover operation uses a simulated binary crossover operator, applying crossover probability to the pairs of individuals formed after the selection operation. Perform parameter recombination and crossover probability. The value is set to 0.9. The mutation operation uses a polynomial mutation operator, applying mutation probabilities to each parameter dimension of the individuals generated by the crossover operation. Perturbation, mutation probability The value is 0.1 at the beginning of the iteration, representing the mutation probability. The mutation probability decreases linearly with the number of iterations, and the mutation probability reaches the preset maximum number of iterations. It dropped to 0.01.

[0046] After each generation of the population completes the fitness value calculation, the optimization solver records the individual with the smallest fitness value in the current generation as the current best individual, and compares its fitness value with the previously recorded best individual, retaining the one with the better fitness value.

[0047] The fitness value calculation, selection, crossover, mutation, and optimal individual update steps are repeatedly executed iteratively until the convergence condition is met. The convergence condition is set as two parallel criteria, and convergence is determined by triggering either one: the first criterion is that the standard deviation of the fitness values ​​of all individuals in the population for three consecutive generations is lower than a preset convergence threshold of 0.001; the second criterion is that the number of iterations reaches a preset maximum number of iterations, which is set to 200. The final retained optimal individual control parameter vector is decoded: the start-up timing encoding segment is mapped sequentially to the start-up time of each agricultural device; the operating power encoding segment is mapped sequentially to the operating power setpoint of each agricultural device; and the movement path deflection angle encoding segment is mapped sequentially to the deflection angle value of each agricultural device at the path intersection point. The set of start-up timing, operating power, and movement path deflection angle is then arranged in chronological order into an optimal linkage control command sequence.

[0048] Example 4: In specific implementation, please refer to Figure 5After the optimal linkage control command sequence is generated, it is split by the command distribution module in the cloud server according to the device identifier. The splitting process, based on the device identifier field carried in the header of each command in the optimal linkage control command sequence, divides the command sequence into several single-device control command sets. Each single-device control command set uniquely corresponds to one agricultural device. The single-device control command set is then arranged in ascending order according to its timestamp, forming the control command queue for that agricultural device.

[0049] The command distribution module sends single-device control commands to the corresponding edge gateways for each agricultural device via the MQTT protocol. The MQTT protocol's Quality of Service (QoS) level is set to QoS1 to ensure that a single-device control command is received by the edge gateway at least once. The topic name used when publishing MQTT messages follows a hierarchical structure of "farm / device / device identifier / control", where "device identifier" is replaced with the unique number registered by the corresponding agricultural device in the production environment. The message payload is encoded in JSON format and includes the following fields: command sequence number, command type, execution timestamp, target operating power value, movement path deflection angle setting value, and command validity period. The command validity period is set to 15 seconds. The basis for this setting is that the maximum interval between two consecutive control commands received by the agricultural device's edge gateway is twice the sampling period. The sampling period is usually set to 5 to 8 seconds. Taking twice the upper limit of the sampling period (8 seconds), i.e., 16 seconds, and rounding down to the nearest integer, we arrive at 15 seconds. If the edge gateway does not receive the next control command within the command validity period, it automatically enters a low-power standby state.

[0050] Upon receiving an MQTT message, the edge gateway of each agricultural device performs JSON parsing on the message payload, extracting the instruction type, execution timestamp, target operating power value, and movement path deflection angle setting. The edge gateway compares the execution timestamp with the current time of its internal clock. When the difference between the two timestamps is less than 50 milliseconds, the edge gateway writes the target operating power value to the power register of the corresponding motor driver and converts the movement path deflection angle setting into the steering motor rotation steps before writing it to the steering controller. Simultaneously, the edge gateway collects the actual operating parameters and environmental factor monitoring values ​​of the agricultural device at a fixed frequency, set to once per second. The actual operating parameters include the actual operating power read from the motor driver and the actual movement path coordinate sequence read from the GPS module. The environmental factor monitoring values ​​include ambient temperature, ambient humidity, and soil moisture content values ​​read from temperature and humidity sensors and soil sensors configured on the agricultural device. The edge gateway encapsulates each collected actual operating parameter and environmental factor monitoring value into a feedback data record. The feedback data record includes a collection timestamp field, a device identifier field, and a parameter key-value pair field, and is published to the cloud server via the MQTT protocol to the topic "farm / device / device identifier number / feedback".

[0051] The cloud server's data receiving module continuously subscribes to the "farm / device / + / feedback" topic. After receiving feedback data records from all agricultural devices, it aligns these records with the acquisition timestamp and the corresponding real-time sensor data stream time window. The alignment method is as follows: the cloud server queries the pre-stored set of original sampled values ​​within the corresponding time window of the real-time sensor data stream. Records in the original sampled value set whose acquisition timestamp differs from the acquisition timestamp of the feedback data record by less than 500 milliseconds are replaced with the parameter key-value pair field content from the feedback data record. This replacement forms the corrected real-time sensor data stream.

[0052] The cloud server invokes the spatiotemporal correlation aggregation processing module, inputting the corrected real-time sensor data stream into it. The module then re-divides the corrected real-time sensor data stream into multiple subsequences according to time windows. For each subsequence, it re-extracts the equipment operating status parameters and environmental factor monitoring values ​​for each agricultural device, recalculates the Pearson correlation coefficients between each pair of agricultural devices, reconstructs the spatial distance weight matrix, and re-executes the weighted superposition and exponential smoothing fusion operations to generate an updated coupling degree matrix. This updated coupling degree matrix replaces the original coupling degree matrix in the cloud storage for use by the next round of optimization solver.

[0053] After the coupling matrix is ​​updated on the cloud server, the deviation analysis module on the cloud server synchronously calculates the deviation vector between the feedback data and the expected parameters in the optimal linkage control command sequence. The deviation vector is constructed as follows: Let the actual value of a parameter of an agricultural device in the feedback data be... The expected parameter value of the corresponding parameter of the agricultural equipment in the optimal linkage control command sequence is... Then the deviation vector component corresponding to this parameter is The deviation vector encompasses the following parameters for all agricultural equipment: operating power, movement path deflection angle, operating torque, and equipment position coordinates. Once constructed, the deviation vector is fed into the online learning module.

[0054] The core architecture of the online learning module is an incremental random forest algorithm, which consists of several decision trees integrated in parallel. Each decision tree adopts a binary regression tree structure. The internal nodes of the binary regression tree store the splitting feature number and splitting threshold, while the leaf nodes store the predicted mean value of the training samples falling into the leaf node. After receiving the bias vector, the incremental random forest algorithm judges the absolute value of each parameter in the bias vector. The threshold is then determined. The specific value is determined by the parameter type: for operating power parameters, the threshold is used for judgment. The threshold is set at 5% of the rated power. This setting is based on the assumption that when the operating power deviation exceeds 5% of the rated power, the energy consumption limit threshold in the multi-device collaborative control constraint model is no longer suitable for the current device's response characteristics. For the movement path deflection angle parameter, a threshold is determined... The setting is 0.05 radians, based on the fact that a deflection angle deviation of 0.05 radians can produce a lateral error of approximately 5 meters over a 100-meter working distance, exceeding the centimeter-level precision tolerance range for precision agriculture operations; for the operating torque parameter, a threshold is determined. Set to 3% of the rated torque; for equipment position coordinate parameters, determine the threshold. Set to 2 meters. When the absolute value of a parameter in the deviation vector exceeds the judgment threshold for the corresponding parameter type. When the online learning module marks the parameter sample as a sample requiring incremental training; when the absolute value of all parameters in the bias vector does not exceed the judgment threshold of the corresponding parameter type. At that time, the online learning module will skip this round of updates.

[0055] For samples marked as requiring incremental training, the online learning module performs the following incremental training steps. Incremental training samples are those whose absolute value in the bias vector exceeds a judgment threshold. The parameters consist of their original values, corresponding expected values, and the values ​​of equipment operation priority weights and energy consumption limit thresholds in the multi-device collaborative control constraint model at the current moment. The online learning module traverses each decision tree in the incremental random forest. For each decision tree, it routes incremental training samples downwards from the root node according to the existing splitting rules until it reaches a leaf node. Upon reaching a leaf node, the online learning module calculates the number of samples after adding new samples. If the number of samples still does not exceed the maximum sample capacity of the leaf node, the new sample is added to the leaf node, and the leaf node's mean prediction is updated. If the number of samples exceeds the maximum sample capacity of the leaf node, a leaf node splitting operation is triggered. The feature that minimizes the mean square error after splitting and the splitting threshold are selected to expand the leaf node into two child nodes. The maximum sample capacity of the leaf node is dynamically adjusted as follows: the frequency of triggering leaf node splits in consecutive incremental training rounds is recorded. When the triggering frequency is higher than the preset upper limit of the splitting frequency, the maximum sample capacity of the leaf node increases to 1.2 times the current value; when the triggering frequency is lower than the preset lower limit of the splitting frequency, the maximum sample capacity of the leaf node decreases to 0.8 times the current value. The preset upper limit for the split frequency is set to trigger 5 times every 10 epochs of incremental training, and the preset lower limit for the split frequency is set to trigger 1 time every 10 epochs of incremental training. Maximum depth of the decision tree. Similarly, the adjustment is dynamic, with the initial value set to 10. Each time a leaf node split is triggered, if the current decision tree depth has reached [a certain value]... Then Add 2.

[0056] After completing this round of incremental training, the online learning module uses the updated incremental random forest to re-predict the equipment operation priority weights and energy consumption limit thresholds in the multi-equipment collaborative control constraint model. During re-prediction, the online learning module constructs a prediction input vector containing the current state statistics of each agricultural device, the average environmental factors within the most recent time window, and the average soil physicochemical indicators. This input vector is then fed into all decision trees of the incremental random forest, and the average of the predicted values ​​of all tree nodes is taken as the output. The output is a multi-dimensional vector, where the first dimension represents the updated suggested equipment operation priority weights, and the second dimension represents the updated suggested energy consumption limit thresholds. The online learning module assigns the updated suggested equipment operation priority weights to the equipment operation priority weight parameters and the updated suggested energy consumption limit thresholds to the energy consumption limit threshold parameters in the multi-equipment collaborative control constraint model, thus completing the parameter update of the multi-equipment collaborative control constraint model.

[0057] Example 5: In specific implementation, please refer to Figure 6After each generation of the population completes the selection process, the optimizer performs a crossover operation on each pair of selected parent individuals. The crossover operation uses a simulated binary crossover operator, which reorganizes the real-number encoded control parameter vector by simulating the behavior of binary-coded crossover.

[0058] The simulated binary crossover operator operates on each dimension of the control parameter vector for the paired parent individuals. For the d-th dimension of the control parameter vector, let the value of the paired parent individual A in that dimension be... The value of the paired parent individual B in this dimension is... The simulated binary crossover operator generates the values ​​of offspring individuals C and D in the d-th dimension. and To complete parameter recombination, the recombination method is determined by the distribution parameters. Control. Distributed parameters The value of determines the degree of dispersion of offspring individuals relative to their parents; distribution parameter The larger the value, the more concentrated the offspring individuals are near the parents; distribution parameter The smaller the value, the further the offspring may deviate from the parents. Distribution parameters The value is set to 20, based on the following: In agricultural equipment collaborative control scenarios, the values ​​of each dimension of the control parameter vector typically exhibit continuous and smooth adaptive terrain characteristics, thus the distributed parameters are taken. To ensure that offspring individuals are sufficiently similar to their parents, the population can maintain its local exploration capability in the search space and avoid skipping the feasible region boundary defined by the constraints due to excessively large search steps.

[0059] When executing the simulated binary crossover operator, for each dimension d of the control parameter vector, a random number following a uniform distribution between 0 and 1 is generated. When random numbers Less than or equal to crossover probability At that time, perform crossover in that dimension to generate the values ​​of offspring individuals C and D in the d-th dimension; when random numbers Greater than the crossover probability At this point, the values ​​of offspring individuals C and D in the d-th dimension directly inherit the values ​​of their parents A and B in that dimension, respectively. (Crossover probability) The value is set to 0.9.

[0060] During the mutation phase, the optimizer performs dimension-by-dimensional mutation on all offspring individuals generated by the crossover operation. The mutation operation employs a multinomial mutation operator, which uses a multinomial distribution to randomly perturb each dimension of the offspring individual's control parameter vector. For the value of the offspring individual in the d-th dimension... The polynomial mutation operator generates new values ​​after mutation. The magnitude of variation is determined by the distribution index. Control. Distribution index. The distribution index determines the amplitude distribution pattern of the variation disturbance. The larger the value, the more concentrated the variation disturbance is around the original value, and the distribution index... The smaller the value, the greater the potential deviation of the variation disturbance. Distribution index The value is set to 30, based on the principle that the variation of control parameters for agricultural equipment must be kept within the physical tolerance range of the equipment and the operational safety boundary, using the distribution index. Setting the value to 30 ensures that over 95% of the variation disturbance amplitude falls within ±15% of the original value, thereby avoiding extreme control parameters that could cause equipment overload or operational trajectory instability due to variation.

[0061] The mutation operation is determined by the mutation probability. Control whether execution occurs in a specific dimension. For each dimension d of the control parameter vector, generate a random number uniformly distributed between 0 and 1. When random number Less than or equal to the probability of mutation When the random number is reached, a polynomial mutation operation is performed on the values ​​of that dimension to generate new mutated values; when the random number is reached... Greater than the probability of mutation When this happens, the value of this dimension remains unchanged.

[0062] Mutation probability It decreases linearly with the number of iterations. At the g-th iteration... mutation probability at the next iteration Calculated using the following formula:

[0063] in, Indicates the first The mutation probability used in the next iteration; Indicates the current iteration number. The value gradually increases from 1 to Positive integers; This indicates the preset maximum number of iterations. The value is set to 200; This represents the initial mutation probability. The value is set to 0.1; This represents the probability of termination mutation. The value is set to 0.01. Initial mutation probability. The rationale for setting it to 0.1 is as follows: In the early stages of the search, the population is highly dispersed, requiring a higher mutation probability to provide sufficient global exploration capability. A mutation probability of 0.1 ensures that approximately 10% of the dimensions in each control parameter vector mutate in each generation, effectively maintaining population diversity. Termination mutation probability The rationale for setting the probability to 0.01 is as follows: Towards the end of the search, the population has concentrated near desirable regions in the solution space. A lower mutation probability avoids excessive disturbance to the converged high-quality solutions. A mutation probability of 0.01 ensures that only a single-digit percentage of the dimensions in the control parameter vector mutates on average, achieving a locally refined search. When the... mutation probability at the next iteration The calculated result is less than the termination mutation probability. When, the probability of mutation Values .

[0064] For dimensions in the control parameter vector that may exceed the legal range of their corresponding parameters after mutation, a boundary reflection repair strategy is adopted: when the mutated values... Exceeding the upper limit of parameter value When this happens, the excess portion will be pushed back into the range of values, and the corrected value will be... When the value after mutation Below the lower limit of parameter value When, the value after repair is If the repaired value still exceeds the limit, the boundary reflection repair strategy is repeated until the value falls within the legal range. The upper and lower limits of each dimension parameter are obtained by transforming the rated power limit of each agricultural device and the geographical boundary of the operating area in the multi-device collaborative control constraint model.

[0065] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A method for optimizing remote linkage and automatic control of agricultural equipment, characterized in that, The method includes: Real-time sensor data streams from multiple agricultural devices within the operating area are collected. These real-time sensor data streams include equipment operating status parameters, environmental factor monitoring values, and soil physicochemical index sequences. The real-time sensing data stream is subjected to spatiotemporal correlation aggregation processing to generate a coupling degree matrix between devices. The coupling degree matrix includes the mutual influence coefficient between devices and the delay response time. A multi-device collaborative control constraint model is constructed based on the coupling degree matrix. The multi-device collaborative control constraint model includes equipment operation priority, energy consumption limit and operation area coverage requirement. The optimization solver is invoked to perform iterative search processing on the multi-device collaborative control constraint model to generate the optimal linkage control command sequence; The optimal linkage control command sequence is distributed to the corresponding devices for execution via a remote communication network, and feedback data is collected in real time to update the coupling matrix.

2. The method for optimizing remote linkage and automatic control of agricultural equipment according to claim 1, characterized in that, The process of performing spatiotemporal correlation aggregation on the real-time sensing data stream to generate a coupling matrix between the devices includes: The real-time sensing data stream is divided into multiple subsequences according to a time window, and each subsequence corresponds to a sampling period. For each subsequence, the operating status parameters and environmental factor monitoring values ​​of each device are extracted, the Pearson correlation coefficient between each pair of devices is calculated, and the spatial distance weight matrix is ​​calculated based on the geographical coordinates of the devices. The Pearson correlation coefficient is weighted and superimposed with the spatial distance weight matrix to generate the inter-device correlation strength matrix for each time window; The correlation strength matrix of multiple consecutive time windows is exponentially smoothed and fused to obtain the coupling degree matrix, wherein the delay response time is determined by the peak position of the cross-correlation function.

3. The method for optimizing remote linkage and automatic control of agricultural equipment according to claim 1, characterized in that, The construction of the multi-device collaborative control constraint model based on the coupling degree matrix includes: From the coupling degree matrix, extract the device pairs whose mutual influence coefficient is greater than a preset threshold, and construct a strongly associated device sub-network graph. The strongly associated device sub-network graph includes node attributes and edge weights. A cooperative controller is assigned to each of the strongly correlated device sub-network graphs, and the cooperative controller maintains the joint state vector of all devices in the sub-network. Based on the equipment operation priority, the optimization weights of each sub-network are allocated. Combined with the energy consumption constraints and operation area coverage requirements, a multi-objective optimization function is established with the goal of minimizing operation completion time and optimizing total energy consumption. The multi-objective optimization function is combined with the delay response time in the coupling degree matrix to form the multi-device collaborative control constraint model.

4. The method for optimizing remote linkage and automatic control of agricultural equipment according to claim 1, characterized in that, The process of calling the optimization solver to iteratively search the multi-device collaborative control constraint model and generate the optimal linkage control command sequence includes: Initialize the population of the optimization solver, with each individual representing a set of control parameter vectors, which include the startup sequence, operating power, and movement path deflection angle of each device; Substitute each individual into the multi-device collaborative control constraint model and calculate the fitness value of that individual. The fitness value is the weighted sum of the multi-objective optimization function and the superposition of the constraint violation penalty term. Based on the fitness value, selection, crossover, and mutation operations are performed on the population to generate a new generation of population, and the current best individual is recorded; Repeat the above steps until the convergence condition is met, and decode the current optimal individual into the optimal linkage control command sequence.

5. The method for optimizing remote linkage and automatic control of agricultural equipment according to claim 4, characterized in that, The population of individuals used to initialize the optimization solver includes: Several sets of control parameters are randomly selected from the historical operation database as the initial parent generation. The historical operation database contains control parameter records of past successful operations. The initial parent generation is subjected to Gaussian perturbation to generate offspring individuals with the same number as the original parent generation. The variance of the Gaussian perturbation is positively correlated with the coefficient of variation of the current environmental factor monitoring values. The initial parent and offspring individuals are merged to form the initial population.

6. The method for optimizing remote linkage and automatic control of agricultural equipment according to claim 1, characterized in that, The step of distributing the optimal linkage control command sequence to corresponding devices for execution via a remote communication network and collecting feedback data in real time to update the coupling matrix includes: The optimal linkage control command sequence is split into single device control commands according to the device identifier and sent to the edge gateway of each device via the MQTT protocol. After parsing the instructions, the edge gateway of each device drives the actuator to perform the action, and at the same time collects the actual operating parameters and environmental factor monitoring values ​​after execution at a fixed frequency as feedback data. The feedback data is uploaded to the cloud server, which replaces the data in the corresponding time window of the real-time sensing data stream with the feedback data, and re-executes the spatiotemporal correlation aggregation process to generate an updated coupling matrix.

7. The method for optimizing remote linkage and automatic control of agricultural equipment according to claim 1, characterized in that, The method further includes, after the real-time acquisition and feedback data: Calculate the deviation vector between the feedback data and the expected parameters in the optimal linkage control command sequence, wherein the deviation vector contains the difference between the actual value and the target value of each parameter; The deviation vector is input into the online learning module, which uses an incremental random forest algorithm to update the parameters in the multi-device collaborative control constraint model. The parameters include the equipment operation priority weight and the energy consumption limit threshold.

8. The method for optimizing remote linkage and automatic control of agricultural equipment according to claim 4, characterized in that, The crossover operation uses a simulated binary crossover operator, and the mutation operation uses a polynomial mutation operator, with the mutation probability decreasing linearly with the number of iterations.

9. The method for optimizing remote linkage and automatic control of agricultural equipment according to claim 7, characterized in that, In the incremental random forest algorithm, each decision tree is incrementally trained only based on parameter samples whose absolute value in the deviation vector exceeds a preset threshold during the update process, and the maximum depth of the tree is dynamically adjusted.

10. A remote linkage and automatic control optimization system for agricultural equipment, characterized in that, The device includes a processor and a memory, the memory and the processor being connected. The memory is used to store programs, instructions or code, and the processor is used to execute the programs, instructions or code in the memory to implement the agricultural equipment remote linkage and automatic control optimization method according to any one of claims 1-9.

Images