High-yield corn planting method and system

Abstract

The invention discloses a corn high-yield planting method and system, and relates to the technical field of corn planting.According to the corn high-yield planting method and system, optimized data with higher pertinence and accuracy are generated through multi-source data fusion and variety specificity parameter weighting, a yield prediction model based on reinforcement learning can process a complex environment and growth interaction relation, and the yield prediction accuracy is improved. More accurate growth prediction data is generated, and a scientific basis is provided for optimization decision making; an optimized planting instruction is generated in combination with a multi-target optimization method of a high-yield target, so that the planting benefit is effectively improved; the automation equipment is controlled to execute an instruction and collect feedback data, so that the operation precision is realized; based on a key event triggering mechanism of execution feedback, model parameters can be dynamically adjusted, subsequent decisions can be continuously optimized, a self-adaptive closed-loop optimization system which continuously learns and improves is formed, and the intelligent level, the management efficiency and the yield potential of corn planting are remarkably improved.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of corn planting, in particular to a corn high-yield planting method and system. BACKGROUND

[0002] Traditional corn planting management relies on farmers' experience and static agronomic knowledge, and it is difficult to cope with the spatial and temporal variability of the field environment and the complexity of the growth dynamics; therefore, a corn planting method that can integrate multi-source data, integrate variety characteristics, and have self-learning and self-adaptive ability needs to be constructed to realize accurate monitoring, intelligent prediction and closed-loop optimization of the whole planting process.

[0003] The prior art such as the invention patent application with the announcement number CN104012269B discloses a corn planting method, which comprises the following steps: land preparation, seed selection, sowing, fertilization, seedling lifting and topdressing. The corn planting method provided by the application has high survival rate, high fertilizer utilization rate, can reduce the use amount of fertilizer, is beneficial to ventilation and the growth of corn, and can effectively increase the yield of corn.

[0004] For the above-mentioned scheme, the inventors of the present application found that the above-mentioned technology at least has the following technical problems: 1. Currently, there is a lack of combination of variety specificity parameters for weighted integration, which cannot be personalized to adapt to the growth characteristics of different corn varieties, cannot generate more accurate and reliable optimization data, and cannot reduce the influence of data noise and deviation; there is a lack of construction of a corn exclusive digital twin model, which cannot accurately simulate the physical process and biological law of corn growth, and cannot realize the mapping of virtual and real; there is a lack of assimilation of real-time data, which cannot dynamically adjust the model parameters, cannot ensure that the model is synchronized with the actual situation in the field, and cannot improve the real-time performance and accuracy of the model.

[0005] 2. Currently, there is a lack of use of reinforcement learning algorithm to process complex nonlinear relationship and dynamic environmental change in corn growth, which cannot adaptively optimize the prediction strategy through trial and error and learning mechanism, cannot improve the accuracy and robustness of yield prediction, cannot identify potential risks in advance, cannot avoid the one-sidedness of single-objective optimization by generating comprehensive optimization planting instructions, and cannot improve the scientificity and feasibility of planting decisions.

[0006] 3. Currently, there is a lack of event triggering mechanism, which cannot timely detect key events and automatically adjust model parameters, cannot form closed-loop optimization to adapt to changing conditions, cannot maintain the optimization effect for a long time, and cannot reduce the planting risk. SUMMARY

[0007] In view of the above-mentioned technical deficiencies, the purpose of the present application is to provide a corn high-yield planting method and system.

[0008] To solve the above technical problems, the application adopts the following technical solutions: In a first aspect, the application provides a high-yield corn planting method, which comprises the following steps: Step one, fusion data generation: based on pre-acquired corn field data, data fusion and corn feature extraction are performed to generate fusion data, and the fusion data is weighted and integrated based on corn variety-specific parameters to obtain optimization data.

[0009] Step two, model state data generation: based on the optimization data, a corn-specific digital twin model is constructed, and real-time data is assimilated to generate model state data.

[0010] Step three, growth prediction: based on the model state data, yield prediction is performed through a reinforcement learning algorithm to generate growth prediction data.

[0011] Step four, optimization instruction generation: based on the growth prediction data, in combination with a high-yield target, optimization planting instructions are generated through multi-objective optimization.

[0012] Step five, optimization planting instruction execution: based on the optimization planting instructions, automatic equipment is controlled to perform operations, and execution feedback data is collected.

[0013] Step six, event-triggered model update: based on the execution feedback data, corn growth key events are monitored, triggering digital twin model parameter adjustment, updating the model state data to form a self-adaptive closed-loop optimization.

[0014] Preferably, the corn field data includes soil temperature data, soil moisture data, plant image data, and corn genome data.

[0015] Preferably, the pre-acquired corn field data is subjected to data fusion and corn feature extraction to generate fusion data, which includes: data cleaning of the corn field data to remove outliers and missing values to generate cleaned data; feature extraction of the cleaned data to extract time series features from soil temperature data and soil moisture data, morphological features from plant image data, and gene expression features from corn genome data to generate a feature data set, and fusion data is generated through a data fusion algorithm.

[0016] Preferably, the fusion data is weighted and integrated based on corn variety-specific parameters to obtain optimization data, which includes: corn variety-specific parameters including photosynthesis gene expression data; based on the corn variety-specific parameters and fusion data, weighted average algorithm is used for integration to generate optimization data.

[0017] Preferably, the constructing a corn-specific digital twin model based on the optimization data and assimilating real-time data to generate model state data comprises: initializing model parameters of the corn growth digital twin model based on the optimization data to obtain an initial model parameter set; integrating corn physiological process algorithms to obtain an integrated physiological model; and then updating the model state according to real-time data corresponding to the model parameters in the initial model parameter set to obtain an updated model state and generate model state data.

[0018] Preferably, the yield prediction based on the model state data through a reinforcement learning algorithm to generate growth prediction data comprises: constructing a reinforcement learning environment, initializing reinforcement learning algorithm parameters, and setting corn growth key stage weights based on the model state data; updating Q values at the same time; and then generating growth prediction data, wherein the growth prediction data comprises yield prediction values and environmental adaptation indicators.

[0019] Preferably, the optimization planting instruction generated by multi-objective optimization based on the growth prediction data in combination with a high yield target comprises: multi-objective optimization calculation of a plurality of optimization objective functions and water and fertilizer dynamic allocation constraint conditions based on the growth prediction data to obtain a corn planting optimal solution set, and extraction of an optimal solution from the corn planting optimal solution set based on the high yield target to generate an optimization planting instruction integrated with water and fertilizer dynamic allocation parameters.

[0020] Preferably, the control of the automatic equipment based on the optimization planting instruction to perform operations and the collection of execution feedback data comprise: analyzing water and fertilizer dynamic allocation parameters based on the optimization planting instruction to generate device control instructions; transmitting the device control instructions to the automatic equipment to control the automatic equipment to perform planting operations; collecting actual growth response data to generate initial feedback data, and performing data cleaning and feature extraction on the initial feedback data to generate execution feedback data.

[0021] Preferably, the monitoring of corn growth key events based on the execution feedback data, the triggering of digital twin model parameter adjustment, the updating of model state data, and the formation of adaptive closed-loop optimization comprise: monitoring corn growth key events through a key event monitoring algorithm based on the execution feedback data to obtain an event flag; triggering digital twin model parameter adjustment when the event flag meets a preset triggering condition, calculating an adjustment parameter set through a parameter optimization algorithm; updating model parameters of the digital twin model based on the adjustment parameter set, and generating updated model state data through a state updating algorithm; and feeding back the updated model state data to step three to form adaptive closed-loop optimization.

[0022] The application provides a corn high-yield planting method in a second aspect, comprising: a fusion data generation module, based on pre-acquired corn field data, data fusion and corn feature extraction are carried out, fusion data is generated, and the fusion data is weighted and integrated based on corn variety-specific parameters to obtain optimization data.

[0023] A model state data generation module, based on the optimization data, constructs a corn-specific digital twin model, and assimilates real-time data to generate model state data.

[0024] A growth prediction module, based on the model state data, carries out yield prediction through a reinforcement learning algorithm to generate growth prediction data.

[0025] An optimization instruction generation module, based on the growth prediction data, combines a high-yield target, and generates optimization planting instructions through multi-objective optimization.

[0026] An optimization planting instruction execution module, based on the optimization planting instructions, controls the automatic equipment to execute operations, and collects execution feedback data.

[0027] An event-triggered model updating module, based on the execution feedback data, monitors corn growth key events, triggers digital twin model parameter adjustment, updates model state data, and forms a self-adaptive closed-loop optimization.

[0028] The application has the following beneficial effects: 1. The corn high-yield planting method and system provided by the application generate more targeted and accurate optimization data through multi-source data fusion and variety-specific parameter weighting. The yield prediction model based on reinforcement learning can handle complex environmental and growth interaction relationships to generate more accurate growth prediction data, providing a scientific basis for optimization decisions. The multi-objective optimization method combined with the high-yield target generates optimization planting instructions, effectively improving planting efficiency. By controlling the automatic equipment to execute instructions and collecting feedback data, the precision of operations is realized. Based on the key event triggering mechanism of execution feedback, the model parameters can be dynamically adjusted to continuously optimize subsequent decisions, forming a self-adaptive closed-loop optimization system that continuously learns and improves, significantly improving the intelligent level, management efficiency, and yield potential of corn planting.

[0029] 2、The application integrates multi-source heterogeneous data through data fusion and feature extraction, overcomes the limitations of single data source, improves the comprehensiveness and consistency of data, and combines with variety-specific parameters for weighted integration, which can individually adapt to the growth characteristics of different corn varieties, generate more accurate and reliable optimization data, lay a high-quality data foundation for subsequent model construction, reduce the influence of data noise and bias, build a corn-specific digital twin model, accurately simulate the physical process and biological law of corn growth, realize the mapping of virtual and reality, and dynamically adjust the model parameters through assimilation of real-time data to ensure that the model is synchronized with the actual field conditions, improve the real-time and accuracy of the model. This makes the model state data more reflect the real growth situation, supports timely decision-making and intervention.

[0030] 3、The application uses reinforcement learning algorithm to process the complex nonlinear relationship and dynamic environmental changes in corn growth, optimizes the prediction strategy through trial and error and learning mechanism, improves the accuracy and robustness of yield prediction. The growth prediction data can identify potential risks in advance; the multi-objective optimization algorithm can balance multiple objectives such as high yield, resource efficiency and environmental sustainability, and generate comprehensive optimized planting instructions (such as irrigation, fertilization, pest control scheme). This avoids the one-sidedness of single-objective optimization, realizes the win-win of economic benefit and ecological benefit, and improves the scientificity and feasibility of planting decision-making.

[0031] 4、The application accurately executes instructions through automatic equipment, reduces human operation errors, improves work efficiency and consistency; at the same time, real-time feedback data is collected to form execution records, providing real data sources for subsequent model verification and updating, ensuring the traceability and controllability of planting operation; through the event triggering mechanism, key events are detected in time and model parameters are automatically adjusted, so that the digital twin model has self-learning and self-adaptive ability; form a closed loop to optimize and adapt to changing conditions, long-term maintain the optimization effect, reduce the risk of planting. BRIEF DESCRIPTION OF DRAWINGS

[0032] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed to be used in the embodiment or prior art description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and those skilled in the art can obtain other drawings according to these drawings without creative labor.

[0033] Figure 1 The method embodiment steps flowchart of the present application.

[0034] Figure 2 The system structure connection diagram of the present application. DETAILED DESCRIPTION

[0035] With reference to the drawings of the embodiments of the present application, the technical solutions in the embodiments of the present application will be clearly and completely described. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by those of ordinary skill in the art without creative work fall within the scope of the present application.

[0036] Please refer to Figure 1 As shown in the drawings, the present application provides a corn high-yield planting method in the first aspect, comprising: step one, fusion data generation: based on the pre-acquired corn field data, data fusion and corn feature extraction are performed to generate fusion data, and the fusion data is weighted and integrated based on corn variety-specific parameters to obtain optimization data.

[0037] In one specific example, the corn field data includes soil temperature data, soil moisture data, plant image data and corn genome data.

[0038] In one specific example, based on the pre-acquired corn field data, data fusion and corn feature extraction are performed to generate fusion data, comprising: data cleaning is performed on the corn field data to remove outliers and missing values to generate cleaned data; feature extraction is performed on the cleaned data to extract time series features from the soil temperature data and the soil moisture data, to extract morphological features from the plant image data, and to extract gene expression features from the corn genome data to generate a feature data set, and fusion data is generated through a data fusion algorithm.

[0039] It should be noted that the data cleaning is performed on the corn field data to remove outliers and missing values to generate cleaned data, and the specific process is as follows: the data cleaning adopts an outlier detection algorithm based on Z-score and a linear interpolation method to handle missing values; wherein the Z-score algorithm calculates the difference between each initial data point and the average value of the data set, and then divides the difference by the standard deviation of the data set to obtain a standardized score; and when the absolute value of the standardized score is greater than 3, the data point is determined to be an outlier and is deleted; wherein the linear interpolation method is used to fill in the missing values, by determining the position of the missing value and the positions and values of the adjacent non-missing values on both sides, calculating the interpolated value based on the linear proportion relationship; the interpolated value is equal to the value of the left adjacent non-missing value plus the difference between the missing value position and the left position, multiplied by the difference between the two values divided by the difference between the two positions. Further, the average value and the standard deviation of the data set are calculated based on historical corn field data.

[0040] It should be noted that the feature extraction is performed on the cleaned data, and the time series features are extracted from the soil temperature and soil moisture data, the morphological features are extracted from the plant image data, and the gene expression features are extracted from the corn genomic data to generate the feature dataset, and the specific process is as follows: the feature extraction includes using a time domain analysis algorithm to extract time series features such as moving average and trend coefficient from the soil temperature and soil moisture data; using an image processing algorithm to extract morphological features such as leaf area index and plant height from the plant image data; using a genomic analysis algorithm to extract gene expression features such as the expression amount of photosynthesis-related genes from the corn genomic data.

[0041] It should be noted that in the time domain analysis algorithm, the calculation method of the moving average is: for a given time point, take the data values of the time point and the previous several time points, add these data values, and then divide by the size of the time window to get the moving average of the time point. The trend coefficient is calculated by a linear regression model, which represents the soil temperature or humidity value as a constant term plus a time variable multiplied by a coefficient, plus an error term; specifically, the soil temperature or humidity value is equal to the intercept term plus the trend coefficient multiplied by the time length, plus the random error. Further, the linear regression model is based on the least squares method to fit the parameters, which determines the intercept term and the trend coefficient by minimizing the sum of the squared differences between the predicted value and the actual value, ensuring the reliability of the trend coefficient.

[0042] It should be noted that in the image processing algorithm, the leaf area index is calculated by image segmentation and pixel statistics, specifically the leaf area index is equal to the total area of leaves in the image divided by the ground area, wherein the total area of leaves is obtained by pixel counting and calibration, and the ground area is calculated by reference scale. The plant height is calculated by proportion, specifically the plant height is equal to the pixel height of the plant in the image divided by the proportion factor, and the proportion factor is determined based on the known reference object in the image to ensure the accuracy of the height measurement.

[0043] It should be noted that in the genomic analysis algorithm, the gene expression features are extracted by a feature selection algorithm, specifically by calculating the importance score of the gene, which is equal to the diagonal elements of the between-class covariance matrix divided by the diagonal elements of the within-class covariance matrix; wherein the between-class covariance matrix measures the difference between different classes, such as the difference between corn varieties, and the within-class covariance matrix measures the variation within the same class; the covariance matrix is calculated based on the gene expression data, and for photosynthesis-related genes, a high importance score indicates that the gene expression has significant discriminability.

[0044] It should be noted that the fusion data is also generated by a data fusion algorithm, wherein the data fusion algorithm uses principal component analysis to reduce the dimensionality and normalize the fusion of the feature dataset to generate the fusion data.

[0045] In one specific example, the weighting integration of the fusion data based on the corn variety-specific parameters to obtain the optimized data includes: the corn variety-specific parameters include photosynthesis gene expression data; and the integration is performed by a weighted average algorithm based on the corn variety-specific parameters and the fusion data to generate the optimized data.

[0046] It should be noted that the integration is performed by a weighted average algorithm based on the corn variety-specific parameters and the fusion data to generate the optimized data, and the specific process is as follows: the weighted integration adopts a weighted average algorithm, each data point in the fusion data is multiplied by a corresponding weight coefficient, and all products are added to obtain the optimized data; wherein the weight coefficient is dynamically adjusted based on the corn variety-specific parameters, for example, the higher the photosynthesis gene expression data, the greater the corresponding weight coefficient. Further, the determination of the weight coefficient is through normalization processing to ensure that the sum of all weight coefficients is 1.

[0047] Step two, model state data generation: based on the optimized data, a corn-specific digital twin model is constructed, and real-time data is assimilated to generate model state data.

[0048] In one specific example, the construction of the corn-specific digital twin model based on the optimized data and the assimilation of real-time data to generate the model state data includes: initializing the model parameters of the corn growth digital twin model based on the optimized data to obtain an initial model parameter set; integrating corn physiological process algorithms to obtain an integrated physiological model; and then updating the model state according to the real-time data corresponding to the model parameters in the initial model parameter set to obtain an updated model state and generate the model state data.

[0049] It should be noted that the initialization of the model parameters of the corn growth digital twin model based on the optimized data to obtain an initial model parameter set has the following specific process: the calculation formula represents the initial value of the th model parameter, represents the value of the th model parameter in the optimized data, represents the weight coefficient of the th model parameter, represents the bias term of the th model parameter; further, the weight coefficient and the bias term are fitted by linear regression based on historical corn growth data to ensure that the initial value of the parameter matches the actual growth conditions; the initial value of the model parameter represents the model parameters such as soil water use efficiency or photosynthesis base rate, the value of the model parameter is the soil humidity or gene expression characteristics in the optimized data, and the weight coefficient and the bias term are used to adjust the parameter scale to make the model more suitable for a specific corn variety.​

[0050] It should be noted that the integrated corn physiological process algorithm is used to obtain the integrated physiological model, and the specific process is as follows: the net photosynthetic rate is divided by the net photosynthetic rate standard value and the photosynthetic rate under light limitation is divided by the photosynthetic rate standard value under light limitation, and is weighted and added, and the dark respiration rate is subtracted from the dark respiration rate standard value; wherein the net photosynthetic rate is represented as the photosynthetic rate under Rubisco enzyme limitation.

[0051] It should be noted that the model state is updated according to the real-time data corresponding to the model parameters in the initial model parameter set to obtain the updated model state, and the specific process is as follows: based on the initial model parameter set, the real-time data corresponding to the model parameters is obtained; and based on the real-time data and the prior model state, the model state is updated by a state update algorithm to obtain the updated model state; based on the updated model state, the model state data is generated by a state output algorithm.

[0052] It should be noted that the calculation formula of the state update algorithm is as follows: In the formula, represents the updated model state, represents the predicted state of the initial model parameter set, represents the Kalman gain matrix, represents the real-time data vector, represents the observation operator matrix; the calculation formula of the Kalman gain matrix is: , wherein represents the prior error covariance matrix, represents the observation error covariance matrix; the predicted state of the initial model parameter set represents the predicted growth state (such as leaf area index) based on the initial model parameter set (such as photosynthesis parameters), the real-time data vector is extracted from the real-time data (such as soil humidity), the Kalman gain matrix adjusts the deviation between the model prediction and the observation, so that the updated model state is closer to the true growth condition, and the observation operator matrix maps the model state to the observation space.

[0053] It should be noted that the observation operator matrix is predefined according to the linear mapping relationship between the state variables and the observation variables. In the Kalman filtering algorithm, the observation operator matrix is used to map the model state vector (such as the growth state of corn) to the observation data space (such as the sensor measurement value), and the elements thereof are determined by analyzing the physical correlation between the state variables and the observation variables; for example, in the corn growth model, if the state variables include leaf area index and biomass, and the observation variable is soil temperature, each row of the observation operator matrix corresponds to an observation variable, and each column corresponds to a state variable.

[0054] It should be noted that the model state data is generated, and the specific process is as follows: the calculation formula is deriving model state data , is represented as a state transition matrix, is represented as an updated model state vector, is represented as a correction constant.

[0055] Further, the model state data such as growth stage index is predefined based on corn growth key indicators (such as photosynthetic efficiency), ensuring that the output data directly reflects the dynamic growth conditions; the model state data is used to quantify the corn growth health degree, for example, a high biomass value indicates suitable growth conditions, facilitating subsequent optimization decisions.

[0056] The present application integrates multi-source heterogeneous data through data fusion and feature extraction, overcomes the limitations of a single data source, and improves the comprehensiveness and consistency of the data; combined with the weighting integration of variety-specific parameters, it can individually adapt to the growth characteristics of different corn varieties, generate more accurate and reliable optimization data, lay a high-quality data foundation for subsequent model construction, and reduce the influence of data noise and bias; the construction of a corn-specific digital twin model can accurately simulate the physical process and biological law of corn growth, realize the mapping of virtual and reality; by assimilating real-time data, dynamically adjusting model parameters, ensuring that the model is synchronized with the actual field conditions, and improving the real-time and accuracy of the model. This makes the model state data more reflect the real growth situation, support timely decision-making and intervention.

[0057] Step three, growth prediction: based on the model state data, yield prediction is performed through a reinforcement learning algorithm to generate growth prediction data.

[0058] In one specific example, the yield prediction based on the model state data through the reinforcement learning algorithm to generate growth prediction data includes: based on the model state data, constructing a reinforcement learning environment, initializing reinforcement learning algorithm parameters, and setting corn growth key stage weights; at the same time, updating the Q value; and then generating growth prediction data, wherein the growth prediction data includes yield prediction value and environmental adaptation index.

[0059] It should be noted that the reinforcement learning environment includes a framework for defining state space, action space, and reward function; wherein the state space includes corn growth indicators, and the action space includes management operation options; the reinforcement learning algorithm parameters include learning rate, discount factor, and exploration rate.

[0060] It should be noted that the reinforcement learning algorithm uses the Q-learning algorithm, and the Q value update formula is as follows:

[0061] wherein, is represented as a time point corn growth status, planting action at time point , reward at time point , calculated by reward function; learning rate, discount factor, corn growth status - planting action at time point corresponding Q value.

[0062] Further, the corn growth status is extracted in the model state data, such as leaf area index or biomass, to represent the corn growth condition; the planting action, such as watering or fertilization operation, is used to simulate the management intervention; the reward is used to evaluate the effect of the planting action; the learning rate is valued in the range of 0 to 1, which is used to control the updating range of the Q value; the discount factor is valued in the range of 0 to 1, which is used to weigh the importance of the current and future rewards; the Q value is used to guide the action selection; wherein the learning and the discount factor are determined by grid search optimization based on historical corn growth data to ensure stable convergence of the algorithm.

[0063] It should be noted that the reward function combines the corn growth key stage weight, and the specific process is as follows: the reward at time point is calculated by the formula , wherein represents the corresponding number of each corn growth key stage, , represents the total number of corn growth key stages, wherein is set to 6, corresponding to the germination stage, seedling stage, jointing stage, tasseling stage, filling stage and maturity stage; represents the weight corresponding to the th corn growth key stage, represents the feature function of the th corn growth key stage, which is used to quantify the growth condition of the stage.

[0064] Further, the weight corresponding to the corn growth key stage is set based on the historical yield data by normalization processing, and the contribution proportion of each stage to the yield is obtained by integral calculation, for example, the filling stage has a higher weight because it has a significant impact on the final yield; the feature function of the corn growth key stage is used to quantify the growth condition of the stage, and the feature function is obtained based on the analysis of the growth index (such as photosynthetic efficiency) in the model state data, to ensure that the reward function fits the physiological process of corn, for example, in the filling stage, the feature function can be calculated as the grain dry weight growth rate.

[0065] It should be noted that the environmental adaptation index is part of the growth prediction data, which is obtained by analyzing the response of the prediction state to the environmental condition (such as drought), and its calculation involves a stress index function; the environmental adaptation index is equal to the predicted yield under the environmental condition divided by the predicted yield under the optimal environmental condition; the index is used to quantify the adaptability of corn to environmental changes, and the higher the value, the stronger the adaptability.

[0066] Further, the calculation of the environmental adaptation index is integrated into the reinforcement learning process, and the predicted output is realized by simulating different environmental scenarios (such as changes in soil moisture), ensuring that the index is practical.

[0067] Step four, optimization instruction generation: based on the growth prediction data, combined with the high yield target, the optimization planting instruction is generated through multi-objective optimization.

[0068] In a specific example, the optimization planting instruction is generated based on the growth prediction data, combined with the high yield target, through multi-objective optimization, including: based on the growth prediction data, a plurality of optimization objective functions and water and fertilizer dynamic allocation constraint conditions are calculated by multi-objective optimization to obtain a set of optimal solutions for corn planting, and based on the high yield target, an optimal solution is extracted from the set of optimal solutions for corn planting to generate an optimization planting instruction integrated with water and fertilizer dynamic allocation parameters.

[0069] It should be noted that the plurality of optimization objective functions includes a yield maximization function and a water use efficiency maximization function; wherein the yield maximization function is constructed based on the yield prediction value in the growth prediction data, and the water use efficiency maximization function is constructed based on the relationship between water input and yield; wherein the water input is irrigation water, and the water use efficiency is the ratio of yield to water input.

[0070] It should be noted that the water and fertilizer dynamic allocation constraint conditions include the upper limit of water resource availability and fertilizer application amount; wherein the water resource availability is the total amount of preset irrigation water source, and the upper limit of fertilizer application amount is set based on soil nutrient demand and environmental standards to prevent over-fertilization.

[0071] It should be noted that based on the plurality of optimization objective functions and water and fertilizer dynamic allocation constraint conditions, the optimization calculation is performed by a multi-objective optimization algorithm to obtain a set of optimal solutions for corn planting; wherein the multi-objective optimization algorithm adopts the weighted sum method, and the specific process is as follows: the comprehensive objective function value is obtained by the calculation formula is represented as a yield maximization function, is a water use efficiency maximization function, and ​​respectively represent a weight factor corresponding to a yield maximization function and a weight factor corresponding to a water use efficiency maximization function; wherein the yield maximization function represents a predicted yield value; the water use efficiency maximization function represents a water use efficiency value.

[0072] Further, , , ; the weight factor corresponding to the yield maximization function and the weight factor corresponding to the water use efficiency maximization function are obtained by factor analysis method, first information condensation of the yield maximization function and the water use efficiency maximization function is performed, then the variance explained rate after rotation is obtained, and the weights are obtained by excluding the cumulative variance explained rate.

[0073] It should be noted that the factor analysis method is a known technology, which is a multivariate statistical analysis method that reduces a number of variables with complex relationships to a few comprehensive factors from the dependent relationship of the internal correlation of the variables; information condensation is represented by calculating the median; the variance explained rate is the amount of information extracted by the factor, and the variance explained rate = eigenvalue / total number of analysis items; the variance explained rate after rotation represents the variance explained rate of the factor after maximum variance rotation.

[0074] It should be noted that the optimal solution set of corn planting is sorted according to the high yield target (such as water use efficiency maximization), and the solution with the highest comprehensive target function value is selected as the optimal solution; further, the irrigation time, irrigation amount, fertilization time point and fertilization amount included in the water and fertilizer dynamic allocation parameters are extracted from the optimal solution of corn planting, and are formatted into executable instructions; for example, the irrigation amount parameter is calculated based on the water allocation value in the optimal solution of corn planting, and the fertilization amount parameter is calculated based on the nutrient allocation value in the optimal solution of corn planting.

[0075] The present application uses reinforcement learning algorithm to process complex nonlinear relationship and dynamic environmental change in corn growth, and through trial and error and learning mechanism, the prediction strategy is self-adaptively optimized to improve the accuracy and robustness of yield prediction. The growth prediction data can identify potential risks in advance; the multi-objective optimization algorithm can balance multiple goals such as high yield, resource efficiency and environmental sustainability, and generate comprehensive optimized planting instructions (such as irrigation, fertilization, pest control scheme). This avoids the one-sidedness of single-objective optimization, realizes the win-win of economic benefit and ecological benefit, and improves the scientificity and feasibility of planting decision.

[0076] Step five, execute the optimized planting instructions: based on the optimized planting instructions, control the automated equipment to perform operations, and collect execution feedback data.

[0077] In one specific example, the controlling the automated device to perform the operation based on the optimized planting instruction and collecting the execution feedback data comprises: analyzing water and fertilizer dynamic allocation parameters based on the optimized planting instruction to generate device control instructions; transmitting the device control instructions to the automated device to control the automated device to perform the planting operation; collecting actual growth response data to generate initial feedback data, and performing data cleaning and feature extraction on the initial feedback data to generate execution feedback data.

[0078] It should be noted that the device control instructions are generated by analyzing water and fertilizer dynamic allocation parameters based on the optimized planting instruction, and the specific process is as follows: the optimized planting instruction includes irrigation time, irrigation amount, fertilization time point and fertilization amount parameters; and the optimized planting instruction is formatted into control signals recognizable by the automated device through an instruction analysis algorithm; wherein the instruction analysis algorithm uses string segmentation and mapping table query to convert text instructions into device protocol codes, for example, the irrigation amount parameter is mapped to a pulse width modulation signal value to drive the unmanned tractor to perform precise irrigation.

[0079] Further, the physical implementation of the instruction analysis algorithm relies on the real-time analysis module of the embedded processor, which converts parameter values into device driving instructions through a lookup table to ensure compatibility of the control signals with the device hardware and improve operation accuracy.

[0080] It should be noted that the device control instructions are transmitted to the automated device to control the automated device to perform the planting operation, and the specific process is as follows: the device control instructions are packaged into data packets based on a wireless communication protocol and transmitted to the automated device through a radio frequency module; after receiving the instructions, the automated device performs the operation through an internal control unit, for example, the unmanned tractor adjusts the nozzle flow rate according to the irrigation amount parameter to complete water and fertilizer application.

[0081] Further, the wireless communication protocol uses LoRa communication technology, which is suitable for farmland environment due to its low power consumption and long distance characteristics, and the data packet packaging includes adding a check code to prevent transmission errors and ensure reliable execution of the instructions.

[0082] It should be noted that the actual growth response data is collected to generate initial feedback data, and the specific process is as follows: the sensor nodes deployed in the field collect soil moisture, leaf temperature and plant image data, and the data is aggregated to the gateway node through multi-hop transmission to generate initial feedback data; wherein the sensor network uses ZigBee protocol networking, and each node periodically samples and sends data.

[0083] Further, the data collection of the sensor nodes is based on a timing trigger mechanism, for example, soil moisture is collected every 30 minutes, and analog signals are converted to digital values through an analog-to-digital converter to ensure real-time and integrity of the data.

[0084] It should be noted that the initial feedback data is subjected to data cleaning and feature extraction to generate execution feedback data, and the specific process is as follows: the data cleaning adopts Z-score-based outlier detection algorithm and moving average filtering method to process noise; wherein the Z-score algorithm obtains a standardized score by subtracting the average value of the data set from the initial feedback data point and then dividing by the standard deviation of the data set; when the absolute value of the standardized score is greater than 3, the initial feedback data point is judged as an outlier and is deleted. The feature extraction uses an image processing algorithm to extract morphological features such as leaf area index and plant height from the plant image data; wherein the leaf area index is calculated by image segmentation and pixel statistics, specifically the leaf area index is equal to the total area of leaves in the image divided by the ground area; the plant height is calculated by proportion, specifically the plant height is equal to the pixel height of the plant in the image divided by the proportion factor.

[0085] Further, the average value and standard deviation of the data set in the Z-score algorithm are calculated based on historical growth response data; the leaf area index calculation in the image processing algorithm relies on color threshold segmentation, identifies green pixels as leaves, and calibrates the ground area by reference scale to improve feature accuracy.

[0086] Step six, event trigger model update: based on the execution feedback data, monitor the key events of corn growth, trigger the parameter adjustment of the digital twin model, update the model state data, and form an adaptive closed-loop optimization.

[0087] In one specific example, the adaptive closed-loop optimization based on the execution feedback data, monitoring the key events of corn growth, triggering the parameter adjustment of the digital twin model, and updating the model state data includes: based on the execution feedback data, monitoring the key events of corn growth by a key event monitoring algorithm to obtain an event flag; when the event flag meets a preset trigger condition, triggering the parameter adjustment of the digital twin model, calculating an adjustment parameter set by a parameter optimization algorithm; based on the adjustment parameter set, updating the model parameters of the digital twin model, and generating updated model state data by a state updating algorithm; feeding back the updated model state data to step three to form an adaptive closed-loop optimization.

[0088] It should be noted that the key event monitoring algorithm is based on plant image data, soil data and environmental data in the execution feedback data, and the event detection is realized by a preset health index calculation and threshold comparison, and the specific process is as follows: the health index is calculated by the formula , , is represented as a morphological feature value based on plant image data, is a soil data feature value, is an environmental data feature value, , and These are represented as the weighting factors corresponding to morphological feature values, soil data feature values, and environmental data feature values.

[0089] Furthermore, the health index is used to quantify the health status of corn growth; morphological characteristics of plant image data, such as leaf area index or leaf color characteristics; soil data characteristics, such as soil moisture or temperature; and environmental data characteristics, such as air temperature or light intensity. Event flags are obtained by comparing the health index with a preset health index threshold; if the health index is less than the preset threshold, the event flag is 1, indicating that a critical event has occurred (such as pests or diseases); otherwise, the event flag is 0, indicating that no event has occurred. The health index threshold is set based on historical pest and disease occurrence data through statistical analysis methods, for example, taking the 10th percentile of the historical health index distribution as the threshold to ensure sensitivity and specificity.

[0090] It should be noted that the parameter optimization algorithm adopts an adaptive adjustment strategy based on event type. It calculates the adjustment parameter set by looking up a preset parameter adjustment mapping table. The specific process is as follows: [Calculation formula...] Determine the set of adjustment parameters , Represented as the parameter vector of the ideal model, This is the current model parameter vector. It is represented as an adjustment coefficient matrix.

[0091] Furthermore, the adjustment parameter set represents the adjustment amount of the digital twin model parameters; the ideal model parameter vector is obtained through regression analysis based on historical optimal growth data; the current model parameter vector is extracted from the digital twin model; the elements of the adjustment coefficient matrix are dynamically determined based on the event type and severity. For example, for pest and disease events, the values ​​of the adjustment coefficient matrix are obtained through least squares optimization based on historical event response data. The physical essence of the parameter optimization algorithm is to access the parameter mapping table in real time through an embedded query module, and retrieve the corresponding adjustment coefficient matrix and current model parameter vector according to the type of event marker (such as pest and disease warning), ensuring that the adjustment parameter set closely matches the actual growth anomaly.

[0092] It should be noted that the implementation of the state update algorithm relies on the real-time assimilation module of the digital twin model. Iterative calculations reduce the deviation between the model and the measured data, making the updated model state data more accurately reflect the actual situation of corn planting. The updated model state data is fed back to the growth prediction step by transmitting it to the reinforcement learning algorithm module through the data bus, and the yield prediction and environmental adaptation index are recalculated, thereby realizing the dynamic optimization of the growth prediction data.

[0093] Further, the essence of adaptive closed-loop optimization is to continuously correct model parameters and states through an event-triggering mechanism, enabling the digital twin model to adapt to changes in the corn growing environment.

[0094] The present application reduces human operation errors and improves work efficiency and consistency by precisely executing instructions through automated equipment; at the same time, real-time execution feedback data is collected to form an execution record, providing a real data source for subsequent model verification and updating, ensuring the traceability and controllability of planting operations; through an event-triggering mechanism, key events are detected in a timely manner and model parameters are automatically adjusted, enabling the digital twin model to have self-learning and adaptive capabilities; a closed-loop optimization is formed to adapt to changing conditions, long-term optimization effects are maintained, and planting risks are reduced.

[0095] Please refer to Figure 2 The present application provides a corn high-yield planting method system in the second aspect.

[0096] The corn high-yield planting method system 100 can be installed in an electronic device. According to the functions implemented, the corn high-yield planting method system 100 can include a fused data generation module 101, a model state data generation module 102, a growth prediction module 103, an optimization instruction generation module 104, an optimized planting instruction execution module 105, and an event-triggered model update module 106. The modules of the present application can also be referred to as units, which refer to a series of computer program segments that can be executed by an electronic device processor and can complete fixed functions, stored in the memory of the electronic device.

[0097] In this embodiment, the functions of each module / unit are as follows: the fused data generation module, based on pre-acquired corn field data, performs data fusion and corn feature extraction, generates fused data, and integrates the fused data based on corn variety-specific parameters to obtain optimized data.

[0098] The model state data generation module, based on the optimized data, constructs a corn-specific digital twin model and assimilates real-time data to generate model state data.

[0099] The growth prediction module, based on the model state data, performs yield prediction through a reinforcement learning algorithm to generate growth prediction data.

[0100] The optimization instruction generation module, based on the growth prediction data, combines a high-yield target to generate optimized planting instructions through multi-objective optimization.

[0101] The optimized planting instruction execution module, based on the optimized planting instructions, controls automated equipment to perform operations and collects execution feedback data.

[0102] An event triggering model updating module monitors a corn growth key event based on the execution feedback data, triggers a digital twin model parameter adjustment, updates model state data, and forms an adaptive closed-loop optimization.

[0103] The corn high-yield planting method and system provided in the application generate more targeted and accurate optimization data through multi-source data fusion and variety-specific parameter weighting. The yield prediction model based on reinforcement learning can process complex environmental and growth interaction relationships to generate more accurate growth prediction data, providing a scientific basis for optimization decisions. The multi-objective optimization method combined with the high-yield target generates optimized planting instructions, effectively improving planting benefits. The implementation of the instructions by the control automation equipment and the collection of feedback data achieve the precision of operations. The key event triggering mechanism based on execution feedback can dynamically adjust model parameters to continuously optimize subsequent decisions, forming an adaptive closed-loop optimization system that continuously learns and improves, significantly improving the intelligent level, management efficiency, and yield potential of corn planting.

[0104] In several embodiments provided in the application, it should be understood that the disclosed methods and systems can be implemented in other ways. For example, the system embodiments described above are only illustrative, and the division of the modules is only a logical functional division. Actual implementation can have another division method.

[0105] The modules described as separate components can or can not be physically separated, and the components displayed as modules can or can not be physical units, i.e., they can be located in one place or distributed to multiple network units. Part or all of the modules can be selected to achieve the purpose of the embodiment according to actual needs.

[0106] In addition, the functional modules in each embodiment of the application can be integrated in one processing unit, or each unit can exist physically, or two or more units can be integrated in one unit. The integrated unit can be realized in the form of hardware or in the form of hardware plus software functional modules.

[0107] It is obvious for those skilled in the art that the application is not limited to the details of the above exemplary embodiments, and the application can be implemented in other specific forms without departing from the spirit or essential characteristics of the application.

[0108] The embodiments of the application can acquire and process related data based on artificial intelligence technology. Artificial intelligence is the use of digital computers or computer-controlled machines to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results.

[0109] It should be pointed out that the above embodiments are only used to illustrate the technical solutions of the present application but not limit the present application. Although the present application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or equivalent replaced without departing from the spirit and scope of the technical solutions of the present application.

Claims

1. A method for high yield planting of corn, characterized by, The method comprises the following steps: Step 1: Fusion data generation: based on the pre-acquired corn field data, data fusion and corn feature extraction are performed to generate fusion data, and the fusion data is weighted and integrated based on corn variety-specific parameters to obtain optimized data; Step 2: Model state data generation: based on the optimized data, a corn-specific digital twin model is constructed, and real-time data is assimilated to generate model state data; Step 3: Growth prediction: based on the model state data, yield prediction is performed through a reinforcement learning algorithm to generate growth prediction data; Step 4: Optimization instruction generation: based on the growth prediction data, high-yield targets are combined to generate optimized planting instructions through multi-objective optimization; Step 5: Execution of optimized planting instructions: based on the optimized planting instructions, automatic equipment is controlled to perform operations, and execution feedback data is collected; Step 6: Event-triggered model updating: based on the execution feedback data, corn growth key events are monitored, triggering parameter adjustment of the digital twin model, updating the model state data, and forming a self-adaptive closed-loop optimization.

2. The method for high yield planting of corn according to claim 1, characterized in that, The corn field data includes soil temperature data, soil moisture data, plant image data, and corn genome data.

3. The method for high yield planting of corn according to claim 1, characterized in that, The data fusion and corn feature extraction based on the pre-acquired corn field data include: Data cleaning is performed on the corn field data to remove outliers and missing values to generate cleaned data; feature extraction is performed on the cleaned data to extract time series features from soil temperature data and soil moisture data, morphological features from plant image data, and gene expression features from corn genome data to generate a feature data set; and fusion data is generated through a data fusion algorithm.

4. The method for high yield planting of corn according to claim 1, characterized in that, The weighted integration of the fusion data based on the corn variety-specific parameters to obtain optimized data includes: The corn variety-specific parameters include photosynthesis gene expression data; the weighted average algorithm is used to integrate the corn variety-specific parameters and the fusion data to generate optimized data.

5. The method for high yield planting of corn according to claim 1, wherein, The construction of the corn-specific digital twin model based on the optimized data and the assimilation of real-time data to generate model state data includes: Based on the optimized data, the model parameters of the corn growth digital twin model are initialized to obtain an initial model parameter set; an integrated physiological model is obtained by integrating corn physiological process algorithms; and then the model state is updated according to the real-time data corresponding to the model parameters in the initial model parameter set to obtain an updated model state, and model state data is generated.

6. The method for high yield planting of corn according to claim 1, wherein, The yield prediction based on the model state data through the reinforcement learning algorithm to generate growth prediction data includes: Based on the model state data, a reinforcement learning environment is constructed, the reinforcement learning algorithm parameters are initialized, and the weights of the key stages of corn growth are set; at the same time, the Q value is updated; and then growth prediction data is generated, which includes yield prediction values and environmental adaptation indicators.

7. The method for high yield planting of corn according to claim 1, wherein, The generation of optimized planting instructions based on the growth prediction data and high-yield targets through multi-objective optimization includes: Based on the growth prediction data, a plurality of optimization target functions and water and fertilizer dynamic allocation constraint conditions are multi-objective optimization calculated to obtain a set of optimal solutions for corn planting, and an optimal solution is extracted from the set of optimal solutions for corn planting based on the high yield target to generate an integrated water and fertilizer dynamic allocation parameter optimization planting instruction.

8. The method for high yield corn planting according to claim 1, characterized in that, Based on the optimization planting instruction, the operation of the automatic equipment is controlled, and execution feedback data is collected, including: Based on the optimization planting instruction, the water and fertilizer dynamic allocation parameters are analyzed, the device control instruction is generated, the device control instruction is transmitted to the automatic equipment to control the automatic equipment to perform planting operation, the actual growth response data is collected, the initial feedback data is generated, and the initial feedback data is data cleaned and feature extracted to generate execution feedback data.

9. The method for high yield planting of corn according to claim 1, wherein, Based on the execution feedback data, the key events of corn growth are monitored, the digital twin model parameter adjustment is triggered, the model state data is updated, and the adaptive closed-loop optimization is formed, including: Based on the execution feedback data, the key events of corn growth are monitored by a key event monitoring algorithm to obtain an event flag; when the event flag meets a preset triggering condition, the digital twin model parameter adjustment is triggered, and an adjustment parameter set is calculated by a parameter optimization algorithm; based on the adjustment parameter set, the model parameters of the digital twin model are updated, and updated model state data is generated by a state updating algorithm; the updated model state data is fed back to step three to form an adaptive closed-loop optimization.

10. A system for performing the method of high yield corn production according to any one of claims 1 to 9, characterized by, Including: The fusion data generation module generates fusion data based on the pre-acquired corn field data, and performs data fusion and corn feature extraction, and integrates the fusion data based on the corn variety specific parameters to obtain optimization data; The model state data generation module constructs a corn exclusive digital twin model based on the optimization data, and assimilates real-time data to generate model state data; The growth prediction module generates growth prediction data by predicting yield based on the model state data through a reinforcement learning algorithm; The optimization instruction generation module generates an optimization planting instruction by multi-objective optimization based on the growth prediction data and in combination with a high yield target; The optimization planting instruction execution module controls the operation of the automatic equipment based on the optimization planting instruction, and collects execution feedback data; The event triggered model updating module monitors the key events of corn growth based on the execution feedback data, triggers the digital twin model parameter adjustment, updates the model state data, and forms the adaptive closed-loop optimization.

Images